Merge dev into fix/native-agent-loop-guard-signals

.env.example

@@ -169,26 +169,6 @@ SEARXNG_INSTANCE=http://localhost:8080
 # ODYSSEUS_STT_MAX_AUDIO_BYTES=26214400                  # speech-to-text audio (25 MB)
 # ODYSSEUS_ICS_MAX_BYTES=10485760                        # calendar .ics import (10 MB)
 
-# ============================================================
-# Host Docker access (explicit opt-in)
-# ============================================================
-# Default Docker Compose does not mount /var/run/docker.sock. Existing
-# Ollama, vLLM, and other OpenAI-compatible endpoints remain usable without it.
-#
-# Enable this only for intentional Cookbook/local Docker-daemon management.
-# Raw socket access is high-trust and can grant broad control over the host
-# Docker daemon. Set DOCKER_GID to the host docker group's numeric GID.
-# Put these values in .env, or export them before running docker compose.
-# COMPOSE_FILE=docker-compose.yml:docker/host-docker.yml
-# DOCKER_GID=963
-# docker/host-docker.yml sets this inside the container. Keep it paired
-# with the socket overlay; setting it alone is not sufficient.
-# ODYSSEUS_ENABLE_HOST_DOCKER=true
-#
-# Host Docker access can be combined with one GPU overlay:
-# COMPOSE_FILE=docker-compose.yml:docker/gpu.nvidia.yml:docker/host-docker.yml
-# COMPOSE_FILE=docker-compose.yml:docker/gpu.amd.yml:docker/host-docker.yml
-
 # ============================================================
 # GPU support (Docker Compose)
 # ============================================================

.github/scripts/focused_test_guidance.py

@@ -1,124 +0,0 @@
-#!/usr/bin/env python3
-"""Report focused pytest guidance for changed paths under tests/."""
-
-from __future__ import annotations
-
-import argparse
-import os
-import shlex
-import subprocess
-import sys
-from collections.abc import Iterable
-from pathlib import PurePosixPath
-
-
-def parse_paths(raw_paths: bytes) -> list[str]:
-    """Decode the NUL-delimited output of ``git diff --name-only -z``."""
-    return [os.fsdecode(path) for path in raw_paths.split(b"\0") if path]
-
-
-def changed_paths_from_merge_base(base_sha: str, head_sha: str) -> list[str]:
-    """Return changed ``tests/`` paths using GitHub PR three-dot semantics.
-
-    GitHub PR changed files are based on the merge base and the PR head, not a
-    direct endpoint diff between the current base branch tip and the PR head.
-    Using the direct endpoint diff can include files changed only on the base
-    branch when the PR branch is stale.
-    """
-    merge_base = subprocess.check_output(
-        ["git", "merge-base", base_sha, head_sha],
-        stderr=subprocess.DEVNULL,
-    ).strip()
-    raw_paths = subprocess.check_output(
-        [
-            "git",
-            "diff",
-            "--name-only",
-            "--diff-filter=ACMRT",
-            "-z",
-            os.fsdecode(merge_base),
-            head_sha,
-            "--",
-            "tests/",
-        ],
-    )
-    return parse_paths(raw_paths)
-
-
-def select_test_paths(paths: Iterable[str]) -> list[str]:
-    """Return unique, repository-relative paths contained by tests/."""
-    selected: set[str] = set()
-    for raw_path in paths:
-        path = PurePosixPath(raw_path)
-        if path.is_absolute() or ".." in path.parts:
-            continue
-        parts = tuple(part for part in path.parts if part != ".")
-        if len(parts) >= 2 and parts[0] == "tests":
-            selected.add(PurePosixPath(*parts).as_posix())
-    return sorted(selected)
-
-
-def is_pytest_file(path: str) -> bool:
-    """Return whether a changed path follows this repository's pytest naming."""
-    name = PurePosixPath(path).name
-    return name.endswith(".py") and (
-        name.startswith("test_") or name.endswith("_test.py")
-    )
-
-
-def pytest_command(paths: Iterable[str]) -> str:
-    """Build a copyable pytest command for changed runnable test files."""
-    command = ["python3", "-m", "pytest", "-q", *paths]
-    return shlex.join(command)
-
-
-def format_report(paths: Iterable[str]) -> str:
-    """Format focused guidance for CI logs and the workflow summary."""
-    changed_paths = select_test_paths(paths)
-    runnable_paths = [path for path in changed_paths if is_pytest_file(path)]
-    lines = ["## Focused test guidance (report-only)", ""]
-    if not changed_paths:
-        lines.append("No changed paths under `tests/`.")
-    else:
-        lines.extend(["Changed paths under `tests/`:", ""])
-        lines.extend(f"- `{path}`" for path in changed_paths)
-    lines.extend(["", "Suggested focused validation:", ""])
-    if runnable_paths:
-        lines.append(f"```sh\n{pytest_command(runnable_paths)}\n```")
-    else:
-        lines.append("No directly runnable pytest files changed.")
-    lines.extend(
-        [
-            "",
-            "This guidance does not infer tests from source changes. "
-            "Existing blocking CI remains the source of truth.",
-        ]
-    )
-    return "\n".join(lines)
-
-
-def _parse_args(argv: list[str]) -> argparse.Namespace:
-    parser = argparse.ArgumentParser(
-        description="Report focused pytest guidance for changed tests/ paths.",
-    )
-    parser.add_argument("--base-sha", help="Pull request base commit SHA.")
-    parser.add_argument("--head-sha", help="Pull request head commit SHA.")
-    return parser.parse_args(argv)
-
-
-def main(argv: list[str] | None = None) -> int:
-    args = _parse_args(sys.argv[1:] if argv is None else argv)
-    if bool(args.base_sha) != bool(args.head_sha):
-        raise SystemExit("--base-sha and --head-sha must be provided together")
-
-    if args.base_sha and args.head_sha:
-        paths = changed_paths_from_merge_base(args.base_sha, args.head_sha)
-    else:
-        paths = parse_paths(sys.stdin.buffer.read())
-
-    print(format_report(paths))
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())

.github/workflows/ci.yml

@@ -15,60 +15,6 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  focused-test-guidance:
-    name: Focused test guidance (report-only)
-    if: github.event_name == 'pull_request'
-    runs-on: ubuntu-latest
-    continue-on-error: true
-    steps:
-      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7.0.0
-        with:
-          fetch-depth: 0
-          persist-credentials: false
-      - name: Report changed test paths
-        env:
-          BASE_SHA: ${{ github.event.pull_request.base.sha }}
-          HEAD_SHA: ${{ github.event.pull_request.head.sha }}
-        run: |
-          report_file="$RUNNER_TEMP/focused-test-guidance.md"
-          publish_report() {
-            cat "$report_file"
-            if [ -n "${GITHUB_STEP_SUMMARY:-}" ]; then
-              cat "$report_file" >> "$GITHUB_STEP_SUMMARY" || true
-            fi
-            return 0
-          }
-
-          report_unavailable() {
-            {
-              printf '%s\n\n' '## Focused test guidance unavailable (report-only)'
-              printf '%s\n\n' "$1"
-              printf '%s\n' 'Existing blocking CI remains the source of truth.'
-            } > "$report_file"
-            publish_report
-            exit 0
-          }
-
-          if [ -z "$BASE_SHA" ] || [ -z "$HEAD_SHA" ]; then
-            report_unavailable "Pull request base/head metadata is missing."
-          fi
-
-          if ! git cat-file -e "${BASE_SHA}^{commit}" 2>/dev/null; then
-            report_unavailable "The pull request base commit is unavailable locally."
-          fi
-
-          if ! git cat-file -e "${HEAD_SHA}^{commit}" 2>/dev/null; then
-            report_unavailable "The pull request head commit is unavailable locally."
-          fi
-
-          if ! python3 .github/scripts/focused_test_guidance.py \
-            --base-sha "$BASE_SHA" \
-            --head-sha "$HEAD_SHA" > "$report_file"; then
-            report_unavailable "The focused test guidance helper could not produce a report."
-          fi
-
-          publish_report
-
   python-syntax:
     name: Python syntax (compileall)
     runs-on: ubuntu-latest

Dockerfile

@@ -32,7 +32,6 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     libgl1 \
     libglib2.0-0t64 \
     libxcb1 \
-    libmagic1 \
     && rm -rf /var/lib/apt/lists/*
 
 # libgl1/libglib2.0-0t64/libxcb1 are runtime shared libs (libGL.so.1,
@@ -41,14 +40,6 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 # and dies with `libxcb.so.1: cannot open shared object file` despite a clean
 # pip install. Using full opencv-python (not -headless) because basicsr/gfpgan/
 # facexlib/realesrgan all depend on the `opencv-python` distribution by name.
-#
-# libmagic1 is the shared lib (libmagic.so.1) that python-magic dlopens for
-# content-based MIME sniffing in src/upload_handler.py. We install both here
-# (libmagic1 + the python-magic wrapper, below) rather than in requirements.txt
-# because python-magic resolves libmagic at import time: where the lib is
-# absent the import can block or raise, so keeping it image-only avoids
-# regressing pip/venv installs on hosts without libmagic. Debian always has the
-# lib here, so the import is instant and detection actually works.
 
 # Docker CLI (client only — daemon stays on the host via the
 # /var/run/docker.sock mount). The Debian `docker.io` package ships
@@ -76,11 +67,6 @@ COPY requirements.txt requirements-optional.txt ./
 RUN pip install --no-cache-dir -r requirements.txt \
     && if [ "$INSTALL_OPTIONAL" = "true" ]; then pip install --no-cache-dir -r requirements-optional.txt; fi
 
-# python-magic powers content-based MIME sniffing in src/upload_handler.py.
-# Image-only (not in requirements.txt) because it needs the libmagic1 system
-# lib installed above; see the apt note near the top of this stage.
-RUN pip install --no-cache-dir python-magic==0.4.27
-
 # Pre-install the patched basicsr/gfpgan/facexlib wheels built in the
 # realesrgan-wheels stage (--no-deps keeps the image lean — torch & friends are
 # pulled only when realesrgan is actually installed). With these dists already

app.py

@@ -3,7 +3,6 @@ import mimetypes
 import os
 import sys
 import asyncio
-import time
 
 # On Windows, asyncio.create_subprocess_exec/shell require the ProactorEventLoop.
 # When started via `python -m uvicorn` from a terminal, uvicorn sets this
@@ -198,50 +197,7 @@ class _RequestTimeoutMiddleware(_BaseHTTPMiddleware):
             )
 
 
-class _InteractiveActivityMiddleware(_BaseHTTPMiddleware):
-    async def dispatch(self, request, call_next):
-        from src.interactive_gate import should_track_interactive_request, track_interactive_request
-
-        path = request.url.path or ""
-        if not should_track_interactive_request(path, request.method):
-            return await call_next(request)
-        async def _stop_background():
-            try:
-                await task_scheduler.stop_background_tasks_for_foreground(reason=f"foreground request {request.method} {path}")
-            except Exception:
-                logging.getLogger("app.foreground_gate").debug("foreground task stop failed", exc_info=True)
-        asyncio.create_task(_stop_background())
-        async with track_interactive_request(path, request.method):
-            return await call_next(request)
-
-
-class _SlowRequestLogMiddleware(_BaseHTTPMiddleware):
-    async def dispatch(self, request, call_next):
-        start = time.perf_counter()
-        status = 500
-        try:
-            response = await call_next(request)
-            status = getattr(response, "status_code", 0) or 0
-            return response
-        finally:
-            elapsed = time.perf_counter() - start
-            try:
-                threshold = float(os.getenv("ODYSSEUS_SLOW_REQUEST_LOG_SECONDS", "0.75") or "0.75")
-            except Exception:
-                threshold = 0.75
-            if elapsed >= threshold:
-                logging.getLogger("app.slow_request").warning(
-                    "slow_request method=%s path=%s status=%s elapsed=%.3fs",
-                    request.method,
-                    request.url.path,
-                    status,
-                    elapsed,
-                )
-
-
 app.add_middleware(_RequestTimeoutMiddleware)
-app.add_middleware(_InteractiveActivityMiddleware)
-app.add_middleware(_SlowRequestLogMiddleware)
 
 # ========= AUTH =========
 from routes.auth_routes import setup_auth_routes, SESSION_COOKIE
@@ -627,20 +583,6 @@ webhook_manager = WebhookManager(api_key_manager=api_key_manager)
 auth_router = setup_auth_routes(auth_manager)
 app.include_router(auth_router)
 
-
-@app.post("/api/activity/heartbeat")
-async def activity_heartbeat():
-    from src.interactive_gate import mark_browser_activity
-    await mark_browser_activity()
-    async def _stop_background():
-        try:
-            await task_scheduler.stop_background_tasks_for_foreground(reason="browser heartbeat")
-        except Exception:
-            logging.getLogger("app.foreground_gate").debug("heartbeat task stop failed", exc_info=True)
-    asyncio.create_task(_stop_background())
-    return {"ok": True}
-
-
 # Uploads
 from routes.upload_routes import setup_upload_routes
 upload_router, upload_cleanup_func = setup_upload_routes(upload_handler)
@@ -662,7 +604,7 @@ from routes.admin_wipe_routes import setup_admin_wipe_routes
 app.include_router(setup_admin_wipe_routes(session_manager))
 
 # Memory
-from routes.memory.memory_routes import setup_memory_routes
+from routes.memory_routes import setup_memory_routes
 memory_router = setup_memory_routes(memory_manager, session_manager, memory_vector=memory_vector)
 app.include_router(memory_router)
 from routes.skills_routes import setup_skills_routes
@@ -679,11 +621,11 @@ app.include_router(setup_chat_routes(
 ))
 
 # Research (background deep-research tasks)
-from routes.research.research_routes import setup_research_routes
+from routes.research_routes import setup_research_routes
 app.include_router(setup_research_routes(research_handler, session_manager=session_manager))
 
 # History
-from routes.history.history_routes import setup_history_routes
+from routes.history_routes import setup_history_routes
 app.include_router(setup_history_routes(session_manager))
 
 # Search
@@ -743,7 +685,7 @@ from routes.signature_routes import setup_signature_routes
 app.include_router(setup_signature_routes())
 
 # Gallery (image library)
-from routes.gallery.gallery_routes import setup_gallery_routes
+from routes.gallery_routes import setup_gallery_routes
 app.include_router(setup_gallery_routes())
 
 # Persisted image-editor drafts (server-backed projects)
@@ -851,7 +793,7 @@ from routes.vault_routes import setup_vault_routes
 app.include_router(setup_vault_routes())
 
 # Contacts (CardDAV)
-from routes.contacts.contacts_routes import setup_contacts_routes
+from routes.contacts_routes import setup_contacts_routes
 app.include_router(setup_contacts_routes())
 
 from companion import setup_companion_routes
@@ -927,34 +869,6 @@ async def get_version():
 async def health_check() -> Dict[str, str]:
     return {"status": "healthy", "timestamp": datetime.now(timezone.utc).isoformat()}
 
-@app.post("/api/client-perf")
-async def client_perf(request: Request):
-    """Low-volume frontend timing reports for stalls that happen before SSE logs."""
-    try:
-        data = await request.json()
-    except Exception:
-        data = {}
-    try:
-        kind = str(data.get("type") or "client").replace("\n", " ")[:80]
-        total_ms = float(data.get("total_ms") or 0)
-        stages = data.get("stages") if isinstance(data.get("stages"), list) else []
-        stage_txt = " ".join(
-            f"{str(s.get('name') or '')[:40]}={float(s.get('delta_ms') or 0):.0f}ms"
-            for s in stages[:20]
-            if isinstance(s, dict)
-        )
-        extra = str(data.get("extra") or "").replace("\n", " ")[:200]
-        logging.getLogger("app.client_perf").warning(
-            "client_perf type=%s total=%.0fms %s%s",
-            kind,
-            total_ms,
-            stage_txt,
-            f" extra={extra}" if extra else "",
-        )
-    except Exception:
-        logging.getLogger("app.client_perf").debug("client_perf log failed", exc_info=True)
-    return {"ok": True}
-
 @app.get("/api/ready")
 async def readiness_check() -> JSONResponse:
     """Readiness / integrity self-check — DB, data dir, local-first storage.
@@ -1051,59 +965,57 @@ async def _startup_event():
 
     _startup_tasks.append(asyncio.create_task(_startup_mcp_connections()))
 
-    # Startup warmups are opt-in. They make later requests a little warmer, but
-    # they also compete with the first seconds of real UI use on slow or busy
-    # machines. Default to clear/idle startup and let requests warm what they use.
-    _startup_warmups_enabled = str(os.getenv("ODYSSEUS_STARTUP_WARMUPS", "")).lower() in {"1", "true", "yes", "on"}
-    if _startup_warmups_enabled:
-        async def _warmup_tool_index():
-            try:
-                from src.tool_index import get_tool_index
-                idx = await asyncio.to_thread(get_tool_index)
-                if idx:
-                    await asyncio.to_thread(idx.get_tools_for_query, "warmup", 8)
-                    logger.info("[startup] Tool index pre-warmed")
-            except Exception as e:
-                logger.warning(f"Tool index warmup failed (non-critical): {type(e).__name__}: {e}")
+    # Pre-warm the RAG tool index off the request path. Loading the local
+    # embedding model + opening ChromaDB + indexing the built-in tools is a
+    # one-time ~1-3s cost that otherwise lands on the user's FIRST message
+    # (showing up as a big `tool_selection` time). Doing it here makes the
+    # first turn as fast as subsequent ones (warm embed ≈ a few ms).
+    async def _warmup_tool_index():
+        try:
+            from src.tool_index import get_tool_index
+            idx = await asyncio.to_thread(get_tool_index)
+            if idx:
+                await asyncio.to_thread(idx.get_tools_for_query, "warmup", 8)
+                logger.info("[startup] Tool index pre-warmed")
+        except Exception as e:
+            logger.warning(f"Tool index warmup failed (non-critical): {type(e).__name__}: {e}")
 
-        _startup_tasks.append(asyncio.create_task(_warmup_tool_index()))
-
-        async def _warmup_endpoints():
-            try:
-                import httpx
-                urls = (
-                    await asyncio.to_thread(model_discovery.warmup_ping_urls)
-                    if model_discovery else []
-                )
-                for url in urls:
-                    try:
-                        async with httpx.AsyncClient(timeout=5.0) as client:
-                            await client.get(url)
-                        logger.info(f"Warmup ping OK: {url}")
-                    except Exception as e:
-                        logger.debug(f"Warmup ping failed for endpoint: {e}")
-            except Exception as e:
-                logger.debug(f"Warmup ping skipped: {e}")
-
-        _startup_tasks.append(asyncio.create_task(_warmup_endpoints()))
-    else:
-        logger.info("Startup warmups disabled (set ODYSSEUS_STARTUP_WARMUPS=1 to enable)")
-
-    # Keep-alive is opt-in. The ping path performs model discovery, and when
-    # stale LAN endpoints are configured it can add periodic backend pressure
-    # that delays unrelated UI requests such as Notes/Documents.
-    _keepalive_enabled = str(os.getenv("ODYSSEUS_MODEL_KEEPALIVE", "")).lower() in {"1", "true", "yes", "on"}
-    if _keepalive_enabled:
-        async def _keepalive_loop():
-            while True:
+    _startup_tasks.append(asyncio.create_task(_warmup_tool_index()))
+    # Warmup: ping all known LLM endpoints to prime connections
+    async def _warmup_endpoints():
+        try:
+            import httpx
+            # model_discovery has no get_endpoints(); that call raised
+            # AttributeError every run and silently disabled warmup/keepalive.
+            # Resolve the /models probe URLs via the real discovery API, off the
+            # event loop since discovery does a blocking port scan.
+            urls = (
+                await asyncio.to_thread(model_discovery.warmup_ping_urls)
+                if model_discovery else []
+            )
+            for url in urls:
                 try:
-                    await asyncio.sleep(60)
-                    await _warmup_endpoints()
+                    async with httpx.AsyncClient(timeout=5.0) as client:
+                        await client.get(url)
+                    logger.info(f"Warmup ping OK: {url}")
                 except Exception as e:
-                    logger.warning(f"Keepalive loop error: {e}")
-                    await asyncio.sleep(300)  # Back off on error
+                    logger.debug(f"Warmup ping failed for endpoint: {e}")
+        except Exception as e:
+            logger.debug(f"Warmup ping skipped: {e}")
 
-        _startup_tasks.append(asyncio.create_task(_keepalive_loop()))
+    _startup_tasks.append(asyncio.create_task(_warmup_endpoints()))
+
+    # Keep-alive: ping endpoints every 60 seconds to prevent cold starts
+    async def _keepalive_loop():
+        while True:
+            try:
+                await asyncio.sleep(60)
+                await _warmup_endpoints()
+            except Exception as e:
+                logger.warning(f"Keepalive loop error: {e}")
+                await asyncio.sleep(300)  # Back off on error
+
+    _startup_tasks.append(asyncio.create_task(_keepalive_loop()))
 
     async def _ensure_default_tasks():
         # Create/reconcile default automation tasks + personal assistant for every user.

core/atomic_io.py

@@ -34,8 +34,6 @@ def atomic_write_json(path: str, data: Any, *, indent: Optional[int] = None) ->
 
 
 def atomic_write_text(path: str, text: str) -> None:
-    if not isinstance(text, str):
-        raise TypeError("atomic_write_text expects a string")
     os.makedirs(os.path.dirname(path) or ".", exist_ok=True)
     tmp = f"{path}.tmp.{os.getpid()}"
     with open(tmp, "w", encoding="utf-8") as f:

core/auth.py

@@ -176,17 +176,16 @@ class AuthManager:
                 )
                 old_user = "admin"
             old_hash = self._config["password_hash"]
-            with self._config_lock:
-                self._config = {
-                    "users": {
-                        old_user: {
-                            "password_hash": old_hash,
-                            "created": time.time(),
-                            "is_admin": True,
-                        }
+            self._config = {
+                "users": {
+                    old_user: {
+                        "password_hash": old_hash,
+                        "created": time.time(),
+                        "is_admin": True,
                     }
                 }
-                self._save()
+            }
+            self._save()
             logger.info(f"Migrated single-user auth to multi-user (admin: {old_user})")
 
     def _drop_reserved_loaded_users(self):
@@ -205,9 +204,8 @@ class AuthManager:
                 continue
             normalized[key] = data
         if removed or normalized != users:
-            with self._config_lock:
-                self._config["users"] = normalized
-                self._save()
+            self._config["users"] = normalized
+            self._save()
         if removed:
             logger.warning(
                 "Removed reserved username(s) from auth config: %s",

core/database.py

@@ -276,7 +276,6 @@ class GalleryImage(TimestampMixin, Base):
     id         = Column(String, primary_key=True, index=True)
     filename   = Column(String, nullable=False, unique=True)
     prompt     = Column(Text, nullable=False, default="")
-    caption    = Column(Text, nullable=True, default="")
     model      = Column(String, nullable=True)
     size       = Column(String, nullable=True)
     quality    = Column(String, nullable=True)
@@ -1183,29 +1182,6 @@ def _migrate_add_multiuser_owner_columns():
     _migrate_add_owner_to_table("documents", "ix_documents_owner")
 
 
-def _migrate_add_gallery_caption_column():
-    """Add OCR/vision caption storage for gallery images."""
-    import sqlite3
-    db_path = DATABASE_URL.replace("sqlite:///", "")
-    if not os.path.exists(db_path):
-        return
-    conn = None
-    try:
-        conn = sqlite3.connect(db_path)
-        columns = [row[1] for row in conn.execute("PRAGMA table_info(gallery_images)").fetchall()]
-        if columns and "caption" not in columns:
-            conn.execute("ALTER TABLE gallery_images ADD COLUMN caption TEXT DEFAULT ''")
-            conn.commit()
-            logging.getLogger(__name__).info("Migrated: added caption column to gallery_images")
-    except Exception as e:
-        logging.getLogger(__name__).warning(f"Migration gallery caption column failed: {e}")
-    finally:
-        try:
-            conn.close()
-        except Exception:
-            pass
-
-
 def _migrate_add_api_token_scopes_column():
     """Add API token scopes for existing installs.
 
@@ -1694,7 +1670,6 @@ class CalendarEvent(TimestampMixin, Base):
     # `Z`-suffix on serialization so the frontend interprets correctly.
     is_utc      = Column(Boolean, default=False, nullable=False)
     rrule       = Column(String, default="")
-    recurrence_exdates = Column(Text, default="")  # JSON list of skipped occurrence starts
     color       = Column(String, nullable=True)  # per-event color override
     status      = Column(String, default="confirmed")  # confirmed, cancelled
     importance  = Column(String, default="normal")    # low | normal | high | critical
@@ -1836,7 +1811,6 @@ def init_db():
     _migrate_add_token_columns()
     _migrate_add_mode_column()
     _migrate_add_multiuser_owner_columns()
-    _migrate_add_gallery_caption_column()
     _migrate_add_api_token_scopes_column()
     _migrate_backfill_document_owner_from_session()
     _migrate_assign_legacy_owner()
@@ -1859,7 +1833,6 @@ def init_db():
     _migrate_add_calendar_origin()
     _migrate_add_calendar_account_id()
     _migrate_add_caldav_sync_columns()
-    _migrate_add_calendar_recurrence_exdates()
     _migrate_chat_messages_fts()
     _migrate_encrypt_email_passwords()
     _migrate_encrypt_signatures()
@@ -2211,28 +2184,6 @@ def _migrate_add_calendar_metadata():
         except Exception:
             pass
 
-
-def _migrate_add_calendar_recurrence_exdates():
-    """Add skipped recurrence occurrences for deleting one instance of a series."""
-    import sqlite3
-    db_path = DATABASE_URL.replace("sqlite:///", "")
-    if not os.path.exists(db_path):
-        return
-    conn = None
-    try:
-        conn = sqlite3.connect(db_path)
-        columns = [row[1] for row in conn.execute("PRAGMA table_info(calendar_events)").fetchall()]
-        if columns and "recurrence_exdates" not in columns:
-            conn.execute("ALTER TABLE calendar_events ADD COLUMN recurrence_exdates TEXT DEFAULT ''")
-        conn.commit()
-    except Exception as e:
-        logging.getLogger(__name__).warning(f"calendar_events recurrence_exdates migration failed: {e}")
-    finally:
-        try:
-            conn.close()
-        except Exception:
-            pass
-
 def get_db():
     """
     Dependency to get a database session.

core/middleware.py

@@ -117,7 +117,7 @@ class SecurityHeadersMiddleware(BaseHTTPMiddleware):
                 f"script-src 'self' 'nonce-{nonce}' https://cdn.jsdelivr.net; "
                 "style-src 'self' 'unsafe-inline' https://cdn.jsdelivr.net; "
                 "font-src 'self' https://cdn.jsdelivr.net; "
-                "img-src 'self' data: blob: https:; "
+                "img-src 'self' data: blob:; "
                 "media-src 'self' blob:; "
                 "connect-src 'self'; "
                 "frame-src 'self'; "

core/session_manager.py

@@ -40,18 +40,7 @@ def _parse_msg_content(raw):
     if isinstance(raw, str) and raw.startswith('[{') and '"type"' in raw:
         try:
             parsed = json.loads(raw)
-            # Only treat as serialized multimodal content when EVERY element is
-            # a dict whose "type" is a recognized content-block kind. Otherwise a
-            # plain text message that merely *looks* like a JSON array of objects
-            # (e.g. a user pasting an API schema/sample with a "type" field) was
-            # silently parsed back into a list, destroying the original string.
-            _BLOCK_TYPES = {
-                "text", "image", "image_url", "audio", "input_audio",
-                "input_image", "document", "file",
-            }
-            if (isinstance(parsed, list) and parsed
-                    and all(isinstance(p, dict) and p.get("type") in _BLOCK_TYPES
-                            for p in parsed)):
+            if isinstance(parsed, list) and all(isinstance(p, dict) for p in parsed):
                 return parsed
         except (json.JSONDecodeError, ValueError):
             pass

docker-compose.gpu-amd.yml

@@ -28,6 +28,14 @@ services:
       # land under /app/.local for the odysseus user. Persist them so a
       # container recreate does not silently remove installed serve engines.
       - ${APP_DATA_DIR:-./data}/local:/app/.local:z
+      # Docker socket — lets Cookbook launch commands like
+      # `docker exec ollama-rocm ollama show <tag>` reach the host's
+      # Docker daemon (and sibling containers like ollama-rocm /
+      # ollama-test). The in-container user needs to be in the
+      # socket's owning group — see `group_add` below; the GID
+      # there must match the host's `docker` group (defaults to 963
+      # on Debian, 999 on Ubuntu — override via env if yours differs).
+      - /var/run/docker.sock:/var/run/docker.sock
     extra_hosts:
       # Lets the container reach local services on the Docker host, including
       # Ollama at http://host.docker.internal:11434.
@@ -93,6 +101,7 @@ services:
       - /dev/kfd
       - /dev/dri
     group_add:
+      - "${DOCKER_GID:-963}"
       - video
       - ${RENDER_GID:-render}
 

docker-compose.gpu-nvidia.yml

@@ -27,6 +27,16 @@ services:
       # land under /app/.local for the odysseus user. Persist them so a
       # container recreate does not silently remove installed serve engines.
       - ${APP_DATA_DIR:-./data}/local:/app/.local:z
+      # Docker socket — lets Cookbook launch commands like
+      # `docker exec ollama-rocm ollama show <tag>` reach the host's
+      # Docker daemon (and sibling containers like ollama-rocm /
+      # ollama-test). The in-container user needs to be in the
+      # socket's owning group — see `group_add` below; the GID
+      # there must match the host's `docker` group (defaults to 963
+      # on Debian, 999 on Ubuntu — override via env if yours differs).
+      - /var/run/docker.sock:/var/run/docker.sock
+    group_add:
+      - "${DOCKER_GID:-963}"
     extra_hosts:
       # Lets the container reach local services on the Docker host, including
       # Ollama at http://host.docker.internal:11434.

docker-compose.yml

@@ -16,6 +16,16 @@ services:
       # land under /app/.local for the odysseus user. Persist them so a
       # container recreate does not silently remove installed serve engines.
       - ${APP_DATA_DIR:-./data}/local:/app/.local:z
+      # Docker socket — lets Cookbook launch commands like
+      # `docker exec ollama-rocm ollama show <tag>` reach the host's
+      # Docker daemon (and sibling containers like ollama-rocm /
+      # ollama-test). The in-container user needs to be in the
+      # socket's owning group — see `group_add` below; the GID
+      # there must match the host's `docker` group (defaults to 963
+      # on Debian, 999 on Ubuntu — override via env if yours differs).
+      - /var/run/docker.sock:/var/run/docker.sock
+    group_add:
+      - "${DOCKER_GID:-963}"
     extra_hosts:
       # Lets the container reach local services on the Docker host, including
       # Ollama at http://host.docker.internal:11434.

docker/entrypoint.sh

@@ -29,12 +29,12 @@ fi
 ODY_USER="$(getent passwd "$PUID" | cut -d: -f1)"
 [ -z "$ODY_USER" ] && ODY_USER=odysseus
 
-# Docker-socket group plumbing for the explicit host-Docker overlay. When
-# opted in, the socket is owned by root:<host docker gid>. Add the app user
-# to that group and later call gosu by username so supplementary groups are
-# retained.
+# Docker-socket group plumbing. When /var/run/docker.sock is bind-mounted
+# (Cookbook uses docker exec to reach sibling containers), the socket is
+# owned by root:<host docker gid>. Add the app user to that group and later
+# call gosu by username so supplementary groups are retained.
 DOCKER_SOCK="${DOCKER_SOCK:-/var/run/docker.sock}"
-if [ "${ODYSSEUS_ENABLE_HOST_DOCKER:-}" = "true" ] && [ -S "$DOCKER_SOCK" ]; then
+if [ -S "$DOCKER_SOCK" ]; then
     SOCK_GID="$(stat -c '%g' "$DOCKER_SOCK" 2>/dev/null || echo '')"
     if [ -n "$SOCK_GID" ] && [ "$SOCK_GID" != "0" ]; then
         if ! getent group "$SOCK_GID" >/dev/null 2>&1; then

docker/host-docker.yml

@@ -1,12 +0,0 @@
-# High-trust host Docker access. Enable only when local Docker-daemon
-# management from Cookbook is required and you accept that raw socket access
-# grants broad control over the host Docker daemon.
-#   COMPOSE_FILE=docker-compose.yml:docker/host-docker.yml
-#   DOCKER_GID=<numeric host Docker group id>
-services:
-  odysseus:
-    volumes:
-      - /var/run/docker.sock:/var/run/docker.sock
-    group_add: ["${DOCKER_GID:-963}"]
-    environment:
-      - ODYSSEUS_ENABLE_HOST_DOCKER=true

docs/setup.md

@@ -99,33 +99,6 @@ Odysseus SSH key and add the public key to the remote server's
 ssh-copy-id -i data/ssh/id_ed25519.pub user@server
 ```
 
-**Host Docker access (explicit opt-in).** Default Docker Compose intentionally
-does not mount `/var/run/docker.sock`. You can still connect Odysseus to
-existing Ollama, vLLM, and other OpenAI-compatible endpoints without Docker
-socket access.
-
-Cookbook/local Docker-daemon management requires the opt-in overlay below. Raw
-Docker socket access is high-trust because it can effectively grant broad
-control over the host Docker daemon. Remote server Docker workflows over SSH
-remain preferred.
-
-Place these values in `.env`, or export them in the shell before running
-`docker compose`:
-
-```bash
-COMPOSE_FILE=docker-compose.yml:docker/host-docker.yml
-DOCKER_GID=<host docker group gid>
-```
-
-Combine host Docker access with a GPU overlay when both are intentionally
-required:
-
-```bash
-COMPOSE_FILE=docker-compose.yml:docker/gpu.nvidia.yml:docker/host-docker.yml
-# or
-COMPOSE_FILE=docker-compose.yml:docker/gpu.amd.yml:docker/host-docker.yml
-```
-
 **Docker GPU overlays.** CPU-only users can skip this section. Cookbook can
 only detect GPUs that Docker exposes to the container — if the host runtime or
 device passthrough is not configured, Cookbook sees the iGPU, another card, or
@@ -326,16 +299,6 @@ To expose Odysseus on a local network or Tailscale with HTTPS:
    ```
 4. Install the `mkcert` CA on any other device you want to access Odysseus from (e.g., for iOS, email the `rootCA.pem` to yourself, install the profile, and trust it in Certificate Trust Settings).
 
-### Common self-host traps (30-second fixes)
-A grab-bag of small gotchas that otherwise turn into long debugging sessions.
-
-- **`AUTH_ENABLED=false` is ignored / you're still forced to log in (Windows).** If you edited `.env` in Notepad it may have saved a UTF-8 **BOM**, turning the first key into `AUTH_ENABLED` so it is never matched. Odysseus loads `.env` with `encoding="utf-8-sig"` to tolerate a leading BOM, but the safe fix is to re-save `.env` as **UTF-8 without BOM** (VS Code: *Save with Encoding → UTF-8*).
-- **macOS: the app isn't at `http://localhost:7000`.** macOS AirPlay Receiver usually holds port `7000`, so the macOS start script serves on **`7860`** instead — open `http://localhost:7860`. To use `7000`, free it (System Settings → General → AirDrop & Handoff → turn off *AirPlay Receiver*) and set `APP_PORT=7000`.
-- **Copy buttons do nothing over a plain-HTTP Tailscale/LAN URL.** Browsers only expose the clipboard API (`navigator.clipboard`) on **secure origins** — HTTPS, or `localhost`. Over `http://100.x.y.z:7860` it is blocked. Serve over HTTPS (see *HTTPS + LAN/Tailscale exposure* above); `localhost` is exempt, so copy still works on the host itself.
-- **Self-hosted ntfy reminders don't reach your phone.** Two things: (1) the bundled ntfy binds to loopback by default — to reach it from your phone set `NTFY_BIND` to your host/Tailscale IP and `NTFY_BASE_URL` to the same server URL in `.env`, then recreate the ntfy container (see the `NTFY_*` block in `.env.example`); (2) in the ntfy **Android** app, subscribe to the topic with **Instant delivery** enabled — non-`ntfy.sh` servers don't get instant push otherwise.
-- **Local mail (Dovecot) login fails: "Plaintext authentication disallowed on non-encrypted connections."** Your IMAP/SMTP server is refusing cleartext auth over an unencrypted link. Prefer enabling TLS on the mail server; on a trusted LAN only, you can allow cleartext (Dovecot: `disable_plaintext_auth = no`).
-- **Calendar/contacts (Radicale) won't sync.** Point Odysseus at the **full collection URL** with its trailing slash — e.g. `http://host:5232/<user>/<collection-id>/` — not just the server root. Radicale shows this address for each calendar/address book in its web UI.
-
 ### Optional Dependencies
 `requirements-optional.txt` contains packages that unlock extra features. It is not installed by default.
 
@@ -472,4 +435,4 @@ All user data lives in `data/` (gitignored): `app.db` (sessions, messages, docum
 `memory.json`, `presets.json`, `uploads/`, `personal_docs/`, `chroma/`, `settings.json`.
 
 To back up or restore everything in `data/`, see the
-[Backup & Restore guide](backup-restore.md).
+[Backup & Restore guide](docs/backup-restore.md).

mcp_servers/email_server.py

@@ -58,11 +58,6 @@ def _uid_fetch_rows(data) -> list:
 _ACCOUNT_CACHE: dict = {}  # key = normalized account selector -> config dict
 _MCP_OWNER_ARG = "_odysseus_owner"
 _CURRENT_OWNER: ContextVar[str | None] = ContextVar("email_mcp_owner", default=None)
-_OWNER_ENV_KEYS = ("ODYSSEUS_MCP_EMAIL_OWNER", "ODYSSEUS_EMAIL_OWNER")
-_OWNER_SCOPE_ERROR = (
-    "Error: email MCP requires an authenticated owner or ODYSSEUS_MCP_EMAIL_OWNER "
-    "when owner-scoped email accounts are configured."
-)
 
 
 def _clean_header_value(value) -> str:
@@ -76,29 +71,13 @@ def _db_path() -> Path:
     return Path(APP_DB)
 
 
-def _configured_owner() -> str | None:
-    for key in _OWNER_ENV_KEYS:
-        owner = os.environ.get(key, "").strip()
-        if owner:
-            return owner
-    return None
-
-
 def _current_owner() -> str:
     owner = _CURRENT_OWNER.get()
-    return str(owner or _configured_owner() or "").strip()
-
-
-def _account_owner(row: dict) -> str:
-    return str(row.get("owner") or "").strip()
-
-
-def _has_owner_scoped_accounts(rows: list[dict]) -> bool:
-    return any(_account_owner(r) for r in rows)
+    return str(owner or "").strip()
 
 
 def _account_visible_to_owner(row: dict, owner: str) -> bool:
-    row_owner = _account_owner(row)
+    row_owner = str(row.get("owner") or "").strip()
     if row_owner == owner:
         return True
     if row_owner:
@@ -117,7 +96,8 @@ def _filter_accounts_for_owner(rows: list[dict]) -> list[dict]:
     if owner:
         return [r for r in rows if _account_visible_to_owner(r, owner)]
 
-    if _has_owner_scoped_accounts(rows):
+    owners = {str(r.get("owner") or "").strip() for r in rows if str(r.get("owner") or "").strip()}
+    if len(owners) > 1:
         return []
     return rows
 
@@ -126,7 +106,8 @@ def _mcp_owner_required(rows: list[dict] | None = None) -> bool:
     if _current_owner():
         return False
     rows = rows if rows is not None else _read_accounts_from_db()
-    return _has_owner_scoped_accounts(rows)
+    owners = {str(r.get("owner") or "").strip() for r in rows if str(r.get("owner") or "").strip()}
+    return len(owners) > 1
 
 
 def _load_email_writing_style() -> str:
@@ -293,8 +274,6 @@ def _load_config(account: str | None = None) -> dict:
     }
 
     raw_rows = _read_accounts_from_db()
-    if _mcp_owner_required(raw_rows):
-        raise ValueError(_OWNER_SCOPE_ERROR)
     rows = _filter_accounts_for_owner(raw_rows)
     row = _resolve_account_from_rows(rows, account)
     if _current_owner() and raw_rows and not rows:
@@ -559,148 +538,6 @@ def _get_cached_summaries():
         return {}
 
 
-def _fixture_email_file() -> Path:
-    return DATA_DIR / "fixture_email_messages.json"
-
-
-def _fixture_email_enabled() -> bool:
-    return _fixture_email_file().exists()
-
-
-def _parse_fixture_date(raw_date: str) -> tuple[str, float]:
-    if not raw_date:
-        return "", 0.0
-    parsed = None
-    try:
-        parsed = datetime.fromisoformat(str(raw_date).replace("Z", "+00:00"))
-    except Exception:
-        try:
-            parsed = email.utils.parsedate_to_datetime(str(raw_date))
-        except Exception:
-            parsed = None
-    if parsed:
-        return parsed.isoformat(), parsed.timestamp()
-    return str(raw_date), 0.0
-
-
-def _fixture_email_record(row: dict, uid_num: int, owner: str) -> dict:
-    sender = str(row.get("from") or "Fixture Sender <fixture@example.invalid>")
-    sender_name, sender_addr = email.utils.parseaddr(sender)
-    date_str, date_epoch = _parse_fixture_date(str(row.get("date") or ""))
-    subject = str(row.get("subject") or "(no subject)")
-    body = str(row.get("body") or "")
-    owner_key = re.sub(r"[^A-Za-z0-9_.-]", "-", owner or "default")
-    uid = str(uid_num)
-    return {
-        "uid": uid,
-        "message_id": f"<fixture-email-{uid}-{owner_key}@fixtures.odysseus.local>",
-        "subject": subject,
-        "from": sender_name or sender_addr or sender,
-        "from_address": sender_addr,
-        "date": date_str,
-        "date_epoch": date_epoch,
-        "summary": body[:240],
-        "body": body,
-        "account": "Fixture Inbox",
-        "account_email": owner or str(row.get("owner") or ""),
-        "account_id": "fixture-email",
-        "attachments": [],
-    }
-
-
-def _fixture_email_rows(owner: str | None = None) -> list[dict]:
-    path = _fixture_email_file()
-    if not path.exists():
-        return []
-    try:
-        raw = json.loads(path.read_text(encoding="utf-8"))
-    except Exception:
-        return []
-    rows = raw.get("messages") if isinstance(raw, dict) else raw
-    out = []
-    owner = str(owner or "").strip()
-    for i, row in enumerate(rows if isinstance(rows, list) else [], start=1):
-        if not isinstance(row, dict):
-            continue
-        row_owner = str(row.get("owner") or "").strip()
-        if owner and row_owner and row_owner != owner:
-            continue
-        out.append(_fixture_email_record(row, i, owner or row_owner))
-    out.sort(key=lambda item: item.get("date_epoch") or 0, reverse=True)
-    return out
-
-
-def _fixture_account_rows() -> list[dict]:
-    if not _fixture_email_enabled():
-        return []
-    owner = _current_owner()
-    owners = []
-    for row in _fixture_email_rows(owner or None):
-        email_addr = row.get("account_email") or owner or "fixture@fixtures.odysseus.local"
-        if email_addr not in owners:
-            owners.append(email_addr)
-    if not owners:
-        owners = [owner or "fixture@fixtures.odysseus.local"]
-    return [
-        {
-            "id": "fixture-email",
-            "owner": owner or owners[0],
-            "name": "Fixture Inbox",
-            "is_default": True,
-            "imap_user": owners[0],
-            "from_address": owners[0],
-        }
-    ]
-
-
-def _fixture_email_matches(item: dict, query: str) -> bool:
-    if not query:
-        return True
-    terms = [term for term in re.split(r"\W+", str(query).lower()) if term]
-    haystack = "\n".join(
-        str(item.get(key) or "")
-        for key in ("subject", "from", "from_address", "body", "summary")
-    ).lower()
-    return all(term in haystack for term in terms)
-
-
-def _fixture_list_emails(folder="INBOX", max_results=20, unresponded_only=False,
-                         unread_only=False, account=None) -> list[dict] | None:
-    if not _fixture_email_enabled():
-        return None
-    if account and str(account).strip().lower() not in {
-        "fixture-email",
-        "fixture inbox",
-        "fixture",
-        str(_current_owner()).lower(),
-    }:
-        return []
-    if (folder or "INBOX").upper() not in {"INBOX", "ALL", "ALL MAIL"}:
-        return []
-    return _fixture_email_rows(_current_owner())[: int(max_results or 20)]
-
-
-def _fixture_search_emails(query, folders=None, max_results=20, account=None) -> list[dict] | None:
-    if not _fixture_email_enabled():
-        return None
-    rows = _fixture_list_emails("INBOX", max_results=1000, account=account) or []
-    out = [dict(row, _folder="INBOX") for row in rows if _fixture_email_matches(row, str(query or ""))]
-    return out[: int(max_results or 20)]
-
-
-def _fixture_read_email(uid=None, message_id=None, folder="INBOX", account=None) -> dict | None:
-    if not _fixture_email_enabled():
-        return None
-    if (folder or "INBOX").upper() not in {"INBOX", "ALL", "ALL MAIL"}:
-        return {"error": f"Email UID {uid or message_id} not found"}
-    for item in _fixture_email_rows(_current_owner()):
-        if uid and str(item.get("uid")) == str(uid):
-            return item
-        if message_id and str(item.get("message_id")) == str(message_id):
-            return item
-    return {"error": f"Email not found with UID/Message-ID: {uid or message_id}"}
-
-
 # ── Tool implementations ──
 
 
@@ -711,9 +548,6 @@ def _list_emails(folder="INBOX", max_results=20, unresponded_only=False,
     Pass unread_only=True and/or unresponded_only=True for attention scans.
     account selects mailbox (None = default).
     """
-    fixture = _fixture_list_emails(folder, max_results, unresponded_only, unread_only, account)
-    if fixture is not None:
-        return fixture
     conn = None
     try:
         conn = _imap_connect(account)
@@ -795,9 +629,6 @@ def _result_sort_time(result: dict) -> datetime:
 
 def _list_emails_across_accounts(folder="INBOX", max_results=20,
                                  unresponded_only=False, unread_only=False):
-    fixture = _fixture_list_emails(folder, max_results, unresponded_only, unread_only, None)
-    if fixture is not None:
-        return fixture, []
     rows = _list_accounts_raw()
     combined = []
     errors = []
@@ -831,9 +662,6 @@ def _search_emails(query, folders=None, max_results=20, account=None):
     _list_emails plus an `_folder` tag."""
     if not query or not str(query).strip():
         return []
-    fixture = _fixture_search_emails(query, folders=folders, max_results=max_results, account=account)
-    if fixture is not None:
-        return fixture
     q = str(query).replace("\\", "\\\\").replace('"', '\\"')
     # Mail clients commonly use OR FROM/SUBJECT/TEXT to match either field.
     # IMAP SEARCH OR is binary, so we nest it.
@@ -956,9 +784,6 @@ def _extract_attachment_to_disk(msg, index, target_dir):
 
 def _read_email(uid=None, message_id=None, folder="INBOX", account=None):
     """Read full email content by UID or message-ID. account = mailbox selector."""
-    fixture = _fixture_read_email(uid=uid, message_id=message_id, folder=folder, account=account)
-    if fixture is not None:
-        return fixture
     cfg = _load_config(account)
     conn = None
     try:
@@ -1012,9 +837,6 @@ def _read_email(uid=None, message_id=None, folder="INBOX", account=None):
 
 
 def _read_email_across_accounts(uid=None, message_id=None, folder="INBOX"):
-    fixture = _fixture_read_email(uid=uid, message_id=message_id, folder=folder, account=None)
-    if fixture is not None:
-        return fixture
     rows = _list_accounts_raw()
     matches = []
     errors = []
@@ -1214,14 +1036,10 @@ def _send_email(to, subject, body, in_reply_to=None, references=None, cc=None, b
     UI. This closes the auto-send hole that let earlier models invent
     signatures and ship them to real recipients without confirmation."""
     if _read_agent_email_confirm_setting():
-        # Even confirmation-first sends must resolve the selected account now.
-        # Otherwise a caller could stage a pending draft against another
-        # owner's account selector before browser approval handles it.
-        cfg = _load_config(account)
         return _stash_agent_draft(
             to=to, subject=subject, body=body,
             in_reply_to=in_reply_to, references=references,
-            cc=cc, bcc=bcc, account=cfg.get("account_id") or account,
+            cc=cc, bcc=bcc, account=account,
         )
     send_account, cfg = _resolve_send_config(account)
     msg = EmailMessage()
@@ -1957,10 +1775,9 @@ async def list_tools() -> list[Tool]:
         Tool(
             name="reply_to_email",
             description=(
-                "Reply to an existing email by UID. This sends immediately. Do NOT use "
-                "for normal 'write/draft a reply saying X' requests; use "
-                "draft_email_reply so the user can review and send from Odysseus. "
-                "Only use this when the user explicitly says to send now. Automatically threads the reply with "
+                "Reply to an existing email by UID. This sends immediately; for normal "
+                "assistant-written replies, prefer draft_email_reply so the user can "
+                "review and send from Odysseus. Automatically threads the reply with "
                 "In-Reply-To and References headers, prefixes 'Re:' on the subject, and "
                 "uses the original sender as the recipient. Set reply_all=true to also CC "
                 "the original To/Cc recipients. For follow-up 'reply ...' requests, use "
@@ -2167,12 +1984,13 @@ async def call_tool(name: str, arguments: dict) -> list[TextContent]:
     try:
         all_db_accounts = _read_accounts_from_db()
         if _mcp_owner_required(all_db_accounts):
-            return [TextContent(type="text", text=_OWNER_SCOPE_ERROR)]
+            return [TextContent(
+                type="text",
+                text="Error: email MCP requires an authenticated owner when multiple email account owners are configured.",
+            )]
 
         if name == "list_email_accounts":
             rows = _filter_accounts_for_owner(all_db_accounts)
-            if not rows:
-                rows = _fixture_account_rows()
             if not rows:
                 if all_db_accounts and owner:
                     return [TextContent(type="text", text="No email accounts configured for this owner.")]

mcp_servers/image_gen_server.py

@@ -73,7 +73,7 @@ async def call_tool(name: str, arguments: dict) -> list[TextContent]:
         if not model_spec:
             for candidate in ("gpt-image-1.5", "gpt-image-1", "dall-e-3"):
                 try:
-                    await asyncio.to_thread(_resolve_model, candidate)
+                    _resolve_model(candidate)
                     model_spec = candidate
                     break
                 except ValueError:
@@ -81,7 +81,7 @@ async def call_tool(name: str, arguments: dict) -> list[TextContent]:
             if not model_spec:
                 return [TextContent(type="text", text="Error: No image model found. Configure one in Admin.")]
 
-        url, model_id, headers = await asyncio.to_thread(_resolve_model, model_spec)
+        url, model_id, headers = _resolve_model(model_spec)
 
         is_gpt_image = "gpt-image" in model_id.lower()
         base_url = url.replace("/chat/completions", "").replace("/v1/messages", "").rstrip("/")

requirements.txt

@@ -3,7 +3,6 @@ uvicorn
 python-multipart
 python-dotenv
 httpx
-httpcore>=1.0,<2.0
 pydantic>=2.13.4
 pydantic-settings>=2.14.1
 SQLAlchemy

routes/calendar_routes.py

@@ -1,7 +1,6 @@
 """Calendar routes — local SQLite-backed calendar CRUD."""
 
 import logging
-import json
 import re
 import uuid
 from datetime import datetime, date, timedelta
@@ -35,24 +34,6 @@ def _ics_naive_dtstart(dt):
         return datetime(dt.year, dt.month, dt.day)
     return dt
 
-
-def _ensure_positive_duration(start_dt, end_dt, all_day):
-    """Clamp an imported event's end so it has a positive duration.
-
-    Some .ics exporters write a single-day all-day event with DTEND equal to
-    DTSTART (treating DTEND as inclusive rather than the RFC 5545 exclusive
-    bound). Stored verbatim that produces a zero-duration row, which the
-    list_events overlap filter (dtstart < end AND dtend > start) silently
-    drops — the event never appears on the calendar even though the web UI
-    would otherwise show it. Normalize a non-positive end to the same default
-    span used when DTEND is absent: one day for all-day events, one hour
-    otherwise.
-    """
-    if end_dt <= start_dt:
-        return start_dt + (timedelta(days=1) if all_day else timedelta(hours=1))
-    return end_dt
-
-
 # Single-user fallback identity. Used only when:
 #   1. The app is configured for single-user (no auth middleware), AND
 #   2. The request didn't resolve to an authenticated user.
@@ -453,20 +434,6 @@ def _parse_dt(s: str) -> datetime:
         if t is not None:
             return base.replace(hour=t[0], minute=t[1])
 
-    # time-first: "3pm today", "9am tomorrow", "11pm tonight"
-    # (parity with parse_due_for_user, which handles these via the same form)
-    m = _re.match(r'^(.+?)\s+(today|tonight|tomorrow|tmrw|yesterday)$', lower)
-    if m:
-        time_part, word = m.group(1).strip(), m.group(2)
-        base = today
-        if word in ("tomorrow", "tmrw"):
-            base = today + timedelta(days=1)
-        elif word == "yesterday":
-            base = today - timedelta(days=1)
-        t = _parse_time(time_part)
-        if t is not None:
-            return base.replace(hour=t[0], minute=t[1])
-
     # next <weekday> [at] TIME
     weekdays = ["monday", "tuesday", "wednesday", "thursday", "friday", "saturday", "sunday"]
     m = _re.match(r'^next\s+(\w+)(?:\s+at)?\s*(.*)$', lower)
@@ -542,7 +509,6 @@ def _event_to_dict(ev: CalendarEvent) -> dict:
         "description": ev.description or "",
         "location": ev.location or "",
         "rrule": ev.rrule or "",
-        "recurrence_exdates": _recurrence_exdates(ev),
         "calendar": ev.calendar.name if ev.calendar else "",
         "calendar_href": ev.calendar_id,
         "color": ev.color or (ev.calendar.color if ev.calendar else ""),
@@ -556,28 +522,6 @@ def _event_to_dict(ev: CalendarEvent) -> dict:
 _RRULE_EXPANSION_LIMIT = 1000
 
 
-def _recurrence_exdates(ev: CalendarEvent) -> list[str]:
-    raw = getattr(ev, "recurrence_exdates", "") or ""
-    if not raw:
-        return []
-    try:
-        values = json.loads(raw)
-    except Exception:
-        return []
-    if not isinstance(values, list):
-        return []
-    return [str(v) for v in values if isinstance(v, str) and v.strip()]
-
-
-def _occurrence_exdate_key(uid: str, ev: CalendarEvent) -> str:
-    if "::" not in uid:
-        return ""
-    suffix = uid.split("::", 1)[1]
-    if ev.all_day:
-        return suffix[:10]
-    return suffix[:16]
-
-
 def _expand_rrule(
     ev: CalendarEvent, start: datetime, end: datetime
 ) -> List[dict]:
@@ -642,7 +586,6 @@ def _expand_rrule(
     results = []
     truncated = False
     base = _event_to_dict(ev)
-    exdates = set(_recurrence_exdates(ev))
 
     for occ_start in rule.xafter(expand_start, inc=True):
         if occ_start >= end:
@@ -663,13 +606,8 @@ def _expand_rrule(
         # Build the compound uid: {base_uid}::{date} or ::{datetime}
         if ev.all_day:
             occ_uid = f"{ev.uid}::{occ_start.strftime('%Y-%m-%d')}"
-            exdate_key = occ_start.strftime("%Y-%m-%d")
         else:
             occ_uid = f"{ev.uid}::{occ_start.strftime('%Y-%m-%dT%H:%M')}"
-            exdate_key = occ_start.strftime("%Y-%m-%dT%H:%M")
-
-        if exdate_key in exdates:
-            continue
 
         d = dict(base)
         d["uid"] = occ_uid
@@ -1180,7 +1118,7 @@ def setup_calendar_routes() -> APIRouter:
             db.close()
 
     @router.delete("/events/{uid}")
-    async def delete_event(request: Request, uid: str, scope: str = "series"):
+    async def delete_event(request: Request, uid: str):
         owner = _require_user(request)
         try:
             base_uid = _resolve_base_uid(uid)
@@ -1189,22 +1127,7 @@ def setup_calendar_routes() -> APIRouter:
         db = SessionLocal()
         try:
             ev = _get_or_404_event(db, base_uid, owner)
-            is_occurrence_delete = scope in {"occurrence", "instance"} and "::" in uid and bool(ev.rrule)
             is_caldav = ev.calendar and ev.calendar.source == "caldav"
-            if is_occurrence_delete:
-                key = _occurrence_exdate_key(uid, ev)
-                if not key:
-                    raise HTTPException(400, "Invalid recurring occurrence uid")
-                exdates = _recurrence_exdates(ev)
-                if key not in exdates:
-                    exdates.append(key)
-                ev.recurrence_exdates = json.dumps(sorted(exdates))
-                if is_caldav:
-                    ev.caldav_sync_pending = "update"
-                db.commit()
-                if is_caldav:
-                    await _push_caldav_event_after_commit(owner, base_uid, "update")
-                return {"ok": True, "scope": "occurrence", "exdate": key}
             if is_caldav:
                 _record_caldav_delete_tombstone(db, ev, owner)
             db.delete(ev)
@@ -1303,7 +1226,7 @@ def setup_calendar_routes() -> APIRouter:
                 db.commit()
                 db.refresh(target_cal)
 
-            imported = skipped = repaired = 0
+            imported = skipped = 0
             for comp in cal_data.walk():
                 if comp.name != "VEVENT":
                     continue
@@ -1339,18 +1262,6 @@ def setup_calendar_routes() -> APIRouter:
                         .first()
                     )
                     if existing:
-                        # An import predating the clamp below may have stored
-                        # this same event with a non-positive duration, which
-                        # the list_events overlap filter hides. Re-importing
-                        # lands here and would skip without touching that row,
-                        # so the event would stay invisible. Backfill the clamp
-                        # onto the stored row before skipping it.
-                        fixed_end = _ensure_positive_duration(
-                            existing.dtstart, existing.dtend, bool(existing.all_day)
-                        )
-                        if fixed_end != existing.dtend:
-                            existing.dtend = fixed_end
-                            repaired += 1
                         skipped += 1
                         continue
 
@@ -1384,8 +1295,6 @@ def setup_calendar_routes() -> APIRouter:
                     else:
                         end_dt = start_dt + timedelta(hours=1)
 
-                end_dt = _ensure_positive_duration(start_dt, end_dt, all_day)
-
                 ev = CalendarEvent(
                     uid=uid_val,
                     calendar_id=target_cal.id,
@@ -1406,7 +1315,6 @@ def setup_calendar_routes() -> APIRouter:
                 "ok": True,
                 "imported": imported,
                 "skipped": skipped,
-                "repaired": repaired,
                 "calendar": cal_display,
                 "calendar_id": target_cal.id,
             }

routes/chat_helpers.py

@@ -14,7 +14,6 @@ from core.database import Session as DBSession, ModelEndpoint
 from src.llm_core import normalize_model_id
 from src.endpoint_resolver import normalize_base
 from src.context_compactor import maybe_compact, trim_for_context
-from src.model_context import estimate_tokens
 from src.auth_helpers import effective_user
 from src.prompt_security import untrusted_context_message
 from routes.prefs_routes import _load_for_user as load_prefs_for_user
@@ -100,19 +99,11 @@ class ChatContext:
     uprefs: dict
     preset: PresetInfo
     preprocessed: PreprocessedMessage
-    context_trimmed: bool = False
-    context_messages_before_trim: int = 0
-    context_messages_after_trim: int = 0
-    context_tokens_before_trim: int = 0
-    context_tokens_after_trim: int = 0
     # Documents auto-created server-side during preprocess (e.g. when an
     # attached fillable PDF gets rendered into a markdown editor doc).
     # The chat route emits a doc_update SSE event for each before streaming
     # begins, so the editor pane switches to the new doc immediately.
     auto_opened_docs: list = field(default_factory=list)
-    # Uploads attached to this user turn, resolved and owner-checked for the
-    # agent's private context. This is not emitted to the browser.
-    uploaded_files: list = field(default_factory=list)
 
 
 # ── Helpers ────────────────────────────────────────────────────────────── #
@@ -375,59 +366,6 @@ async def preprocess(
     )
 
 
-def build_uploaded_file_manifest(att_ids: list, upload_handler, owner: Optional[str]) -> list[dict]:
-    """Resolve current-turn upload IDs into a small tool-facing manifest.
-
-    The chat UI already sends attachment ids, and preprocessing inlines as much
-    text as fits. Agent mode still needs a discoverable bridge for files whose
-    content was truncated/omitted or when the model chooses file tools. Only
-    owner-authorized uploads are included, and paths must remain inside the
-    configured upload directory.
-    """
-    if not att_ids or not upload_handler or not hasattr(upload_handler, "resolve_upload"):
-        return []
-
-    def _read_file_can_open(path: str) -> bool:
-        try:
-            from src.tool_execution import _resolve_tool_path
-
-            return _resolve_tool_path(path) == os.path.realpath(path)
-        except Exception:
-            return False
-
-    manifest: list[dict] = []
-    for att_id in att_ids:
-        try:
-            info = upload_handler.resolve_upload(str(att_id), owner=owner)
-        except Exception:
-            logger.debug("Failed to resolve upload %r for agent manifest", att_id, exc_info=True)
-            continue
-        if not isinstance(info, dict):
-            continue
-
-        path = info.get("path")
-        if path:
-            try:
-                inside = True
-                if hasattr(upload_handler, "_inside_upload_dir"):
-                    inside = bool(upload_handler._inside_upload_dir(path))
-                elif hasattr(upload_handler, "inside_base_dir"):
-                    inside = bool(upload_handler.inside_base_dir(path))
-                if not inside or not os.path.exists(path) or not _read_file_can_open(path):
-                    path = None
-            except Exception:
-                path = None
-
-        manifest.append({
-            "id": info.get("id") or str(att_id),
-            "name": info.get("name") or info.get("original_name") or str(att_id),
-            "mime": info.get("mime", ""),
-            "size": info.get("size", 0),
-            "path": path,
-        })
-    return manifest
-
-
 def add_user_message(sess, chat_handler, preprocessed: PreprocessedMessage, incognito: bool = False):
     """Add user message to session history and update session name.
     In incognito mode, still add to in-memory history (for conversation context)
@@ -675,11 +613,6 @@ async def build_chat_context(
     # bearer-token chat requests use the token owner instead of the "api" sentinel.
     user = effective_user(request)
     uprefs = load_prefs_for_user(user)
-    uploaded_files = build_uploaded_file_manifest(
-        att_ids or [],
-        getattr(chat_handler, "upload_handler", None),
-        getattr(sess, "owner", None),
-    )
     casual_low_signal = _is_casual_low_signal(message)
 
     # Memory enabled?
@@ -783,12 +716,7 @@ async def build_chat_context(
     messages, context_length, was_compacted = await maybe_compact(
         sess, sess.endpoint_url, sess.model, messages, sess.headers, owner=user,
     )
-    _before_trim_messages = len(messages)
-    _before_trim_tokens = estimate_tokens(messages)
     messages = trim_for_context(messages, context_length)
-    _after_trim_messages = len(messages)
-    _after_trim_tokens = estimate_tokens(messages)
-    _context_trimmed = _after_trim_messages < _before_trim_messages or _after_trim_tokens < _before_trim_tokens
 
     return ChatContext(
         preface=preface,
@@ -802,13 +730,7 @@ async def build_chat_context(
         uprefs=uprefs,
         preset=preset,
         preprocessed=preprocessed,
-        context_trimmed=_context_trimmed,
-        context_messages_before_trim=_before_trim_messages,
-        context_messages_after_trim=_after_trim_messages,
-        context_tokens_before_trim=_before_trim_tokens,
-        context_tokens_after_trim=_after_trim_tokens,
         auto_opened_docs=auto_opened_docs,
-        uploaded_files=uploaded_files,
     )
 
 

routes/chat_routes.py

@@ -3,7 +3,6 @@
 import asyncio
 import json
 import os
-import re
 import time
 import logging
 from datetime import datetime
@@ -41,13 +40,8 @@ from routes.chat_helpers import (
     clean_thinking_for_save,
     _enforce_chat_privileges,
 )
-from src.action_intents import ToolIntent, classify_tool_intent as _classify_tool_intent
-from src.tool_policy import (
-    WEB_TOOL_NAMES,
-    build_effective_tool_policy,
-    is_web_search_explicitly_denied,
-    web_search_enabled_for_turn,
-)
+from src.action_intents import classify_tool_intent as _classify_tool_intent
+from src.tool_policy import build_effective_tool_policy
 
 logger = logging.getLogger(__name__)
 
@@ -69,78 +63,6 @@ def _stream_set(session_id: str, **fields) -> None:
     rec.update(fields)
 
 
-def _message_plain_text(content: Any) -> str:
-    if isinstance(content, list):
-        parts: List[str] = []
-        for block in content:
-            if isinstance(block, dict):
-                text = block.get("text")
-                if isinstance(text, str):
-                    parts.append(text)
-            elif isinstance(block, str):
-                parts.append(block)
-        return " ".join(parts)
-    return str(content or "")
-
-
-def _last_user_plain_text(messages: List[Dict[str, Any]]) -> str:
-    for msg in reversed(messages or []):
-        if msg.get("role") == "user":
-            return _message_plain_text(msg.get("content"))
-    return ""
-
-
-def _ensure_current_request_is_latest_user(messages: List[Dict[str, Any]], current_message: str) -> List[Dict[str, Any]]:
-    """Defensively keep detached streams grounded on the request that created them."""
-    current = str(current_message or "").strip()
-    if not current:
-        return messages
-    latest = _last_user_plain_text(messages).strip()
-    if latest == current or current in latest or latest in current:
-        return messages
-    logger.warning(
-        "[chat_stream] latest user context mismatch; appending current request for model call. latest=%r current=%r",
-        latest[:120],
-        current[:120],
-    )
-    repaired = list(messages or [])
-    repaired.append({"role": "user", "content": current})
-    return repaired
-
-
-_WEB_FOLLOWUP_RE = re.compile(
-    r"^\s*(?:(?:can|could|would|will)\s+you\s+)?"
-    r"(?:check|try\s+again|look(?:\s+now|\s+it\s+up)?|search(?:\s+now|\s+online|\s+it)?|"
-    r"do\s+it|again)\??\s*$",
-    re.I,
-)
-_RECENT_WEB_CONTEXT_RE = re.compile(
-    r"\b(?:weather|forecast|rain|raining|hourly|news|headlines|rate|exchange|currency|"
-    r"price|current|latest|search|look\s+up|online)\b",
-    re.I,
-)
-
-
-def _recent_session_text(sess, limit: int = 8, max_chars: int = 2000) -> str:
-    history = getattr(sess, "history", None) or getattr(sess, "_history", None) or []
-    chunks: List[str] = []
-    for msg in history[-limit:]:
-        content = getattr(msg, "content", None)
-        if content is None and isinstance(msg, dict):
-            content = msg.get("content")
-        text = _message_plain_text(content).strip()
-        if text:
-            chunks.append(text)
-    return " ".join(chunks)[-max_chars:]
-
-
-def _is_contextual_web_followup(message: str, sess) -> bool:
-    """Treat short retry/check replies as web lookups when recent context was web."""
-    if not message or not _WEB_FOLLOWUP_RE.search(message):
-        return False
-    return bool(_RECENT_WEB_CONTEXT_RE.search(_recent_session_text(sess)))
-
-
 def _resolve_request_workspace(request, raw_value) -> tuple:
     """Resolve the posted workspace for this request: (workspace, rejected).
 
@@ -588,7 +510,6 @@ def setup_chat_routes(
         # below). Skill extraction should only learn from real agent sessions,
         # not chats we quietly promoted for a notes/calendar intent.
         user_requested_agent = (chat_mode == "agent")
-        _search_enabled = web_search_enabled_for_turn(allow_web_search, use_web)
         # Intent auto-escalation: if the user is clearly asking the assistant
         # to create a todo, reminder, or calendar event, promote chat → agent
         # for this turn so the LLM has access to manage_notes / manage_calendar.
@@ -606,10 +527,6 @@ def setup_chat_routes(
                 _tool_intent.category,
                 _tool_intent.reason,
             )
-        elif chat_mode == "chat" and _search_enabled:
-            chat_mode = "agent"
-            auto_escalated = True
-            logger.info("chat→agent auto-escalation: search enabled")
         active_doc_id = form_data.get("active_doc_id", "").strip()
         logger.info(f"[doc-inject] chat_mode={chat_mode}, active_doc_id={active_doc_id!r}")
 
@@ -702,20 +619,6 @@ def setup_chat_routes(
                     400,
                     "No model selected for this chat. Open the model picker and choose one before sending.",
                 )
-            if (
-                chat_mode == "chat"
-                and isinstance(message, str)
-                and (not _tool_intent or not _tool_intent.needs_tools)
-                and _is_contextual_web_followup(message, sess)
-            ):
-                _tool_intent = ToolIntent(True, "web", "contextual web lookup follow-up")
-                chat_mode = "agent"
-                auto_escalated = True
-                logger.info(
-                    "chat→agent auto-escalation: category=%s reason=%s",
-                    _tool_intent.category,
-                    _tool_intent.reason,
-                )
         except SessionNotFoundError as e:
             raise HTTPException(404, str(e))
         except (ValueError, ValidationError):
@@ -826,15 +729,6 @@ def setup_chat_routes(
                         logger.info(f"[doc-inject] found by ID: title={active_doc.title!r}, lang={active_doc.language!r}, is_active={active_doc.is_active}, content_len={len(active_doc.current_content or '')}")
                 else:
                     logger.warning(f"[doc-inject] NOT FOUND by ID {active_doc_id}")
-            if not active_doc:
-                _email_doc_q = _doc_db.query(DBDocument).filter(
-                    DBDocument.session_id == session,
-                    DBDocument.is_active == True,
-                    DBDocument.language == "email",
-                )
-                active_doc = _owner_session_filter(_email_doc_q, ctx.user).order_by(DBDocument.updated_at.desc()).first()
-                if active_doc:
-                    logger.info(f"[doc-inject] found email draft by session fallback: title={active_doc.title!r}")
             if not active_doc:
                 _session_doc_q = _doc_db.query(DBDocument).filter(
                     DBDocument.session_id == session,
@@ -872,35 +766,20 @@ def setup_chat_routes(
 
         # Build disabled-tools set from frontend toggles + user privileges
         disabled_tools = set()
-        # Only disable bash when the caller *explicitly* set it to a falsy
-        # value. When unset (None), defer to per-user privilege checks below.
-        # Web search is per-turn opt-in: either the chat pre-search setting
-        # (`use_web=true`) or agent web toggle (`allow_web_search=true`) must
-        # explicitly enable it.
+        # Only disable bash/web_search when the caller *explicitly* set them
+        # to a falsy value.  When unset (None), defer to per-user privilege
+        # checks below — this lets admins with can_use_bash=True use bash
+        # by default without having to send allow_bash in every request.
         if allow_bash is not None and str(allow_bash).lower() != "true":
             disabled_tools.add("bash")
         _explicit_web_intent = bool(_tool_intent and _tool_intent.category == "web")
-        if is_web_search_explicitly_denied(allow_web_search) or not _search_enabled:
-            disabled_tools.update(WEB_TOOL_NAMES)
-        if _explicit_web_intent:
-            # A direct lookup/search request should not drift into personal
-            # tools or shell fallbacks. It can only use web_search/web_fetch
-            # when the request's explicit web setting enabled them.
-            disabled_tools.update({
-                "bash", "python",
-                "search_chats", "manage_skills", "manage_memory",
-                "read_file", "write_file", "edit_file",
-                "create_document", "edit_document", "update_document",
-                "send_email", "reply_to_email",
-                "manage_notes", "manage_calendar", "manage_tasks",
-                "api_call", "builtin_browser",
-            })
-            if _search_enabled:
-                disabled_tools.difference_update(WEB_TOOL_NAMES)
-            else:
-                disabled_tools.update(WEB_TOOL_NAMES)
-        elif _search_enabled:
-            disabled_tools.difference_update(WEB_TOOL_NAMES)
+        if (
+            allow_web_search is not None
+            and str(allow_web_search).lower() != "true"
+            and not _explicit_web_intent
+        ):
+            disabled_tools.add("web_search")
+            disabled_tools.add("web_fetch")
 
         # Nobody/incognito mode: deny tools that would expose the user's
         # persistent memory, past chats, or other identity-linked data.
@@ -911,19 +790,19 @@ def setup_chat_routes(
                 "manage_skills",      # skill presets tied to user
             })
 
-        # Active email reader open → strip the tools that let the agent drift
-        # away from the visible email or skip review. The only allowed compose
-        # path is ui_control open_email_reply, which opens the same draft editor
-        # as the Reply button with the generated body pre-filled. This prevents
-        # the model from falling back to direct SMTP when it botches a draft
-        # call, and prevents fake email-shaped documents.
+        # Active email reader open → strip the tools that let the agent
+        # "drift" to a new compose: create_document (writes a fake email-
+        # shaped .md file) and send_email (sends fresh to a recipient the
+        # agent invented). With those gone, the only paths left for "write
+        # email saying X" are ui_control open_email_reply (draft) and
+        # reply_to_email (immediate send) — both of which use the open
+        # email's UID. Code-level enforcement instead of relying on a
+        # prompt rule the model can ignore.
         if active_email_ctx and active_email_ctx.get("uid"):
             disabled_tools.update({
                 "create_document",
                 "send_email",
-                "reply_to_email",
                 "mcp__email__send_email",
-                "mcp__email__reply_to_email",
             })
 
         # Enforce per-user privileges
@@ -951,7 +830,11 @@ def setup_chat_routes(
         from src.settings import get_setting
         _global_disabled = get_setting("disabled_tools", [])
         if _global_disabled and isinstance(_global_disabled, list):
-            disabled_tools.update(_global_disabled)
+            explicit_web_allowed = allow_web_search is not None and str(allow_web_search).lower() == "true"
+            if explicit_web_allowed:
+                disabled_tools.update(t for t in _global_disabled if t not in {"web_search", "web_fetch"})
+            else:
+                disabled_tools.update(_global_disabled)
 
         # Light auto-escalation: the user is in chat mode and just expressed a
         # notes/calendar/email intent. Grant the relevant managers but withhold
@@ -1165,16 +1048,13 @@ def setup_chat_routes(
                     _active_streams.pop(session, None)
                     return
 
-            messages = _ensure_current_request_is_latest_user(ctx.messages, message)
+            messages = ctx.messages
 
             # Auto-compact notification
             if ctx.was_compacted:
                 yield f"data: {json.dumps({'type': 'compacted', 'context_length': ctx.context_length})}\n\n"
-            if ctx.context_trimmed and not ctx.was_compacted:
-                yield f"data: {json.dumps({'type': 'context_trimmed', 'data': {'context_length': ctx.context_length, 'messages_before': ctx.context_messages_before_trim, 'messages_after': ctx.context_messages_after_trim, 'tokens_before': ctx.context_tokens_before_trim, 'tokens_after': ctx.context_tokens_after_trim}})}\n\n"
 
             full_response = ""
-            thinking_response = ""
             last_metrics = None
 
             # Configured fallback chain for the default chat model. Tried in
@@ -1264,9 +1144,7 @@ def setup_chat_routes(
                                     # Forward them so the client can show a thinking
                                     # indicator, but don't fold them into the saved
                                     # reply (mirrors the rewrite path below).
-                                    if data.get("thinking"):
-                                        thinking_response += data["delta"]
-                                    else:
+                                    if not data.get("thinking"):
                                         full_response += data["delta"]
                                         _stream_set(session, partial=full_response)
                                     yield chunk
@@ -1286,12 +1164,6 @@ def setup_chat_routes(
                                     _reported_model = last_metrics.get("model")
                                     last_metrics["requested_model"] = _requested_model
                                     last_metrics["model"] = _reported_model or _actual_model or _answered_by or _requested_model
-                                    if ctx.context_trimmed:
-                                        last_metrics["context_trimmed"] = True
-                                        last_metrics["context_messages_before_trim"] = ctx.context_messages_before_trim
-                                        last_metrics["context_messages_after_trim"] = ctx.context_messages_after_trim
-                                        last_metrics["context_tokens_before_trim"] = ctx.context_tokens_before_trim
-                                        last_metrics["context_tokens_after_trim"] = ctx.context_tokens_after_trim
                                     if ctx.context_length and last_metrics.get("input_tokens"):
                                         pct = min(round((last_metrics["input_tokens"] / ctx.context_length) * 100, 1), 100.0)
                                         last_metrics["context_percent"] = pct
@@ -1334,11 +1206,8 @@ def setup_chat_routes(
                                 }
                                 yield f'data: {json.dumps({"type": "metrics", "data": last_metrics})}\n\n'
                             if full_response:
-                                _metrics_to_save = dict(last_metrics or {})
-                                if thinking_response.strip() and not _metrics_to_save.get("thinking"):
-                                    _metrics_to_save["thinking"] = thinking_response.strip()
                                 _saved_id = save_assistant_response(
-                                    sess, session_manager, session, full_response, _metrics_to_save,
+                                    sess, session_manager, session, full_response, last_metrics,
                                     character_name=ctx.preset.character_name,
                                     web_sources=web_sources,
                                     rag_sources=ctx.rag_sources,
@@ -1351,7 +1220,7 @@ def setup_chat_routes(
                                     yield f'data: {json.dumps({"type": "message_saved", "id": _saved_id})}\n\n'
                                 run_post_response_tasks(
                                     sess, session_manager, session, message, full_response,
-                                    _metrics_to_save, ctx.uprefs, memory_manager, memory_vector, webhook_manager,
+                                    last_metrics, ctx.uprefs, memory_manager, memory_vector, webhook_manager,
                                     incognito=incognito, compare_mode=compare_mode,
                                     character_name=ctx.preset.character_name,
                                     owner=_user,
@@ -1386,14 +1255,7 @@ def setup_chat_routes(
                 try:
                     from src.settings import get_setting
                     from src.agent_tools import MAX_AGENT_ROUNDS as _DEFAULT_ROUNDS
-                    # Per-message tool budget from settings; guard defensively in
-                    # case settings.json was hand-edited to a non-numeric value
-                    # (the HTTP admin endpoint validates, but direct edits bypass
-                    # it). 0 = unlimited, matching auth_routes set_settings().
-                    try:
-                        _tool_budget = int(get_setting("agent_max_tool_calls", 0))
-                    except (TypeError, ValueError):
-                        _tool_budget = 0
+                    _tool_budget = int(get_setting("agent_max_tool_calls", 0))
                     # Per-message round cap from settings; clamp defensively in
                     # case settings.json was hand-edited to a bad value.
                     try:
@@ -1403,8 +1265,8 @@ def setup_chat_routes(
                     _max_rounds = max(1, min(_max_rounds, 200))
 
                     _forced_tools = None
-                    if _search_enabled:
-                        _forced_tools = set(WEB_TOOL_NAMES)
+                    if allow_web_search is not None and str(allow_web_search).lower() == "true":
+                        _forced_tools = {"web_search", "web_fetch"}
 
                     async for chunk in stream_agent_loop(
                         sess.endpoint_url,
@@ -1428,7 +1290,6 @@ def setup_chat_routes(
                         approved_plan=approved_plan or None,
                         workspace=workspace or None,
                         forced_tools=_forced_tools,
-                        uploaded_files=ctx.uploaded_files,
                     ):
                         if chunk.startswith("data: ") and not chunk.startswith("data: [DONE]"):
                             try:
@@ -1437,9 +1298,7 @@ def setup_chat_routes(
                                     # Reasoning tokens arrive flagged thinking:true.
                                     # Forward them for the live indicator, but keep
                                     # them out of the saved reply (same as chat mode).
-                                    if data.get("thinking"):
-                                        thinking_response += data["delta"]
-                                    else:
+                                    if not data.get("thinking"):
                                         full_response += data["delta"]
                                         _stream_set(session, partial=full_response)
                                     yield chunk
@@ -1451,6 +1310,8 @@ def setup_chat_routes(
                                     "doc_stream_open", "doc_stream_delta",
                                     "doc_update", "doc_suggestions", "ui_control",
                                     "rounds_exhausted",
+                                    "loop_breaker_triggered",
+                                    "intent_nudge_exhausted",
                                     "ask_user",
                                     "plan_update",
                                 ):
@@ -1477,26 +1338,15 @@ def setup_chat_routes(
                                     _reported_model = last_metrics.get("model")
                                     last_metrics["requested_model"] = last_metrics.get("requested_model") or _requested_model
                                     last_metrics["model"] = _reported_model or _actual_model or _answered_by or _requested_model
-                                    if ctx.context_trimmed:
-                                        last_metrics["context_trimmed"] = True
-                                        last_metrics["context_messages_before_trim"] = ctx.context_messages_before_trim
-                                        last_metrics["context_messages_after_trim"] = ctx.context_messages_after_trim
-                                        last_metrics["context_tokens_before_trim"] = ctx.context_tokens_before_trim
-                                        last_metrics["context_tokens_after_trim"] = ctx.context_tokens_after_trim
                                     yield f'data: {json.dumps({"type": "metrics", "data": last_metrics})}\n\n'
                             except json.JSONDecodeError:
                                 yield chunk
                         elif chunk.startswith("event: "):
                             yield chunk
                         elif chunk == "data: [DONE]\n\n":
-                            _has_tool_events = bool((last_metrics or {}).get("tool_events"))
-                            if full_response or _has_tool_events:
-                                _response_to_save = full_response or "Done."
-                                _metrics_to_save = dict(last_metrics or {})
-                                if thinking_response.strip() and not _metrics_to_save.get("thinking"):
-                                    _metrics_to_save["thinking"] = thinking_response.strip()
+                            if full_response:
                                 _saved_id = save_assistant_response(
-                                    sess, session_manager, session, _response_to_save, _metrics_to_save,
+                                    sess, session_manager, session, full_response, last_metrics,
                                     character_name=ctx.preset.character_name,
                                     web_sources=web_sources,
                                     rag_sources=ctx.rag_sources,
@@ -1506,8 +1356,8 @@ def setup_chat_routes(
                                 if _saved_id:
                                     yield f'data: {json.dumps({"type": "message_saved", "id": _saved_id})}\n\n'
                                 run_post_response_tasks(
-                                    sess, session_manager, session, message, _response_to_save,
-                                    _metrics_to_save, ctx.uprefs, memory_manager, memory_vector, webhook_manager,
+                                    sess, session_manager, session, message, full_response,
+                                    last_metrics, ctx.uprefs, memory_manager, memory_vector, webhook_manager,
                                     incognito=incognito, compare_mode=compare_mode,
                                     character_name=ctx.preset.character_name,
                                                             agent_rounds=_agent_rounds,

routes/codex_routes.py

@@ -15,7 +15,6 @@ from typing import Any
 from fastapi import APIRouter, BackgroundTasks, Body, HTTPException, Request
 from fastapi.responses import StreamingResponse
 
-from core.middleware import require_admin
 from src.auth_helpers import require_authenticated_request, require_user
 from src.tool_implementations import do_manage_notes
 from src.constants import COOKBOOK_STATE_FILE
@@ -110,20 +109,6 @@ def _scope_owner_all(request: Request, required: set[str]) -> str:
     return require_user(request)
 
 
-def _require_cookbook_scope(request: Request, allowed: set[str]) -> str:
-    """Authorize a Codex cookbook route.
-
-    For API-token callers, enforce the given scope set.
-    For cookie-session callers, additionally require admin privileges
-    because cookbook surfaces expose host topology, task logs, tmux
-    commands, and model-serving controls.
-    """
-    owner = _scope_owner(request, allowed)
-    if not getattr(request.state, "api_token", False):
-        require_admin(request)
-    return owner
-
-
 def _find_endpoint(router: APIRouter | None, method: str, path: str):
     if router is None:
         return None
@@ -133,18 +118,6 @@ def _find_endpoint(router: APIRouter | None, method: str, path: str):
     return None
 
 
-def _clamp_pagination(offset: Any, limit: Any, *, default_limit: int = 50, max_limit: int = 50) -> tuple[int, int]:
-    try:
-        parsed_offset = int(0 if offset in (None, "") else offset)
-    except (TypeError, ValueError):
-        raise HTTPException(400, "Invalid offset")
-    try:
-        parsed_limit = int(default_limit if limit in (None, "") else limit)
-    except (TypeError, ValueError):
-        raise HTTPException(400, "Invalid limit")
-    return max(0, parsed_offset), max(1, min(parsed_limit, max_limit))
-
-
 def setup_codex_routes(
     email_router: APIRouter | None = None,
     memory_router: APIRouter | None = None,
@@ -452,18 +425,10 @@ def setup_codex_routes(
         owner = _scope_owner(request, DOCS_READ_SCOPES)
         if documents_library_endpoint is None:
             raise HTTPException(503, "Documents integration is not available")
-        offset, limit = _clamp_pagination(offset, limit)
-        result = await _as_owner(
+        return await _as_owner(
             request, owner, documents_library_endpoint,
             request, search, language, sort, offset, limit, archived,
         )
-        if isinstance(result, dict):
-            docs = result.get("documents")
-            total = result.get("total")
-            if isinstance(docs, list) and isinstance(total, int):
-                next_offset = offset + len(docs)
-                result["next_offset"] = next_offset if next_offset < total else None
-        return result
 
     @router.get("/documents/{doc_id}")
     async def codex_documents_get(request: Request, doc_id: str):
@@ -567,14 +532,14 @@ def setup_codex_routes(
 
     @router.get("/cookbook/tasks")
     async def codex_cookbook_tasks(request: Request):
-        _require_cookbook_scope(request, COOKBOOK_READ_SCOPES)
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
         state = _read_cookbook_state()
         tasks = state.get("tasks") or []
         return {"tasks": [_redact_task(t) for t in tasks]}
 
     @router.get("/cookbook/servers")
     async def codex_cookbook_servers(request: Request):
-        _require_cookbook_scope(request, COOKBOOK_READ_SCOPES)
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
         state = _read_cookbook_state()
         servers = state.get("env", {}).get("servers") or []
         # Strip ssh creds / passwords; keep only what's needed to pick a host.
@@ -593,7 +558,7 @@ def setup_codex_routes(
 
     @router.get("/cookbook/output/{session_id}")
     async def codex_cookbook_output(request: Request, session_id: str, tail: int = 400):
-        _require_cookbook_scope(request, COOKBOOK_READ_SCOPES)
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
         # Defensive: session_id must be the tmux-style id we issue
         # (`serve-XXXX` / `cookbook-XXXX` / `queue-XXXX`); anything else
         # would let the agent run arbitrary `tmux capture-pane` targets.
@@ -635,7 +600,7 @@ def setup_codex_routes(
 
     @router.post("/cookbook/serve")
     async def codex_cookbook_serve(request: Request, body: dict[str, Any] = Body(default_factory=dict)):
-        _require_cookbook_scope(request, COOKBOOK_LAUNCH_SCOPES)
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
         # Wraps /api/model/serve with the SAME validation the UI uses.
         # _validate_serve_cmd (called inside model_serve) rejects shell
         # metachars and requires the leading binary to be in the
@@ -674,7 +639,7 @@ def setup_codex_routes(
 
     @router.post("/cookbook/stop/{session_id}")
     async def codex_cookbook_stop(request: Request, session_id: str):
-        _require_cookbook_scope(request, COOKBOOK_LAUNCH_SCOPES)
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
         import re as _re
         if not _re.fullmatch(r"[a-zA-Z0-9_-]+", session_id):
             raise HTTPException(400, "Invalid session id")
@@ -694,7 +659,7 @@ def setup_codex_routes(
         """List cached models on a configured server (or local if host is omitted).
         Mirrors `list_cached_models` from the chat agent so external agents have
         the same inventory view before deciding what to serve/download."""
-        _require_cookbook_scope(request, COOKBOOK_READ_SCOPES)
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
         # Hit /api/model/cached internally, with the same modelDirs the chat
         # agent's list_cached_models would resolve from cookbook state.
         state = _read_cookbook_state()
@@ -756,7 +721,7 @@ def setup_codex_routes(
         """List saved serve presets (model + host + port + launch cmd).
         Counterpart to `list_serve_presets`. Use BEFORE composing a `serve`
         body — the user's saved preset usually has the working cmd already."""
-        _require_cookbook_scope(request, COOKBOOK_READ_SCOPES)
+        _scope_owner(request, COOKBOOK_READ_SCOPES)
         state = _read_cookbook_state()
         presets = state.get("presets") or []
         out = []
@@ -776,7 +741,7 @@ def setup_codex_routes(
     async def codex_cookbook_serve_preset(request: Request, name: str):
         """Launch a saved preset by name. Reuses the working cmd + host the
         user already saved, avoiding the cmd-allowlist trial-and-error loop."""
-        _require_cookbook_scope(request, COOKBOOK_LAUNCH_SCOPES)
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
         import re as _re
         if not _re.fullmatch(r"[A-Za-z0-9 _.:@\-]+", name):
             raise HTTPException(400, "Invalid preset name")
@@ -828,7 +793,7 @@ def setup_codex_routes(
         cookbook tracking. Needed when serve_model rejects a cmd and the
         agent falls back to direct ssh — without adoption the session is
         invisible to the UI. Body: {tmux_session, model, host?, port?}."""
-        _require_cookbook_scope(request, COOKBOOK_LAUNCH_SCOPES)
+        _scope_owner(request, COOKBOOK_LAUNCH_SCOPES)
         norm = dict(body or {})
         sess = (norm.get("tmux_session") or norm.get("session_id") or "").strip()
         model = (norm.get("model") or norm.get("repo_id") or "").strip()

routes/contacts/__init__.py

@@ -1,5 +0,0 @@
-"""Contacts route domain package (slice 2e, #4082/#4071).
-
-Contains contacts_routes.py, migrated from the flat routes/ directory.
-Backward-compat shim at routes/contacts_routes.py re-exports from here.
-"""

routes/contacts/contacts_routes.py

@@ -1,916 +0,0 @@
-"""
-contacts_routes.py
-
-CardDAV contacts integration. Reads from local Radicale, supports
-search and adding new contacts.
-"""
-
-import re
-import logging
-import uuid
-import json
-import csv
-import io
-import os
-import inspect
-import httpx
-from pathlib import Path
-from datetime import datetime
-from urllib.parse import urljoin, urlparse, urlunparse
-
-from core.log_safety import redact_url
-from fastapi import APIRouter, Query, Depends, Response, HTTPException
-from typing import List, Dict, Optional
-
-from core.middleware import require_admin
-from src.url_safety import check_outbound_url
-
-logger = logging.getLogger(__name__)
-
-from src.constants import DATA_DIR as _DATA_DIR, SETTINGS_FILE as _SETTINGS_FILE, CONTACTS_FILE as _CONTACTS_FILE
-DATA_DIR = Path(_DATA_DIR)
-SETTINGS_FILE = Path(_SETTINGS_FILE)
-LOCAL_CONTACTS_FILE = Path(_CONTACTS_FILE)
-
-
-def _load_settings():
-    if SETTINGS_FILE.exists():
-        return json.loads(SETTINGS_FILE.read_text(encoding="utf-8"))
-    return {}
-
-
-def _save_settings(settings):
-    from core.atomic_io import atomic_write_json
-    atomic_write_json(str(SETTINGS_FILE), settings, indent=2)
-
-
-def _get_carddav_config():
-    import os
-    settings = _load_settings()
-    password = settings.get("carddav_password", os.environ.get("CARDDAV_PASSWORD", ""))
-    if password and "carddav_password" in settings:
-        from src.secret_storage import decrypt
-        password = decrypt(password)
-    return {
-        "url": settings.get("carddav_url", os.environ.get("CARDDAV_URL", "")),
-        "username": settings.get("carddav_username", os.environ.get("CARDDAV_USERNAME", "")),
-        "password": password,
-    }
-
-
-def _carddav_configured(cfg: Optional[Dict] = None) -> bool:
-    cfg = cfg or _get_carddav_config()
-    return bool((cfg.get("url") or "").strip())
-
-
-def _validate_carddav_url(url: str) -> str:
-    cleaned = (url if isinstance(url, str) else "").strip().rstrip("/")
-    ok, reason = check_outbound_url(
-        cleaned,
-        block_private=os.getenv("CARDDAV_BLOCK_PRIVATE_IPS", "false").lower() == "true",
-    )
-    if not ok:
-        raise ValueError(f"Rejected CardDAV URL: {reason}")
-    return cleaned
-
-
-def _carddav_base_url(cfg: Dict) -> str:
-    return _validate_carddav_url(cfg.get("url") or "")
-
-
-def _normalize_contact(contact: Dict) -> Dict:
-    emails = []
-    for e in contact.get("emails") or ([] if not contact.get("email") else [contact.get("email")]):
-        e = str(e or "").strip()
-        if e and e not in emails:
-            emails.append(e)
-    phones = []
-    for p in contact.get("phones") or ([] if not contact.get("phone") else [contact.get("phone")]):
-        p = str(p or "").strip()
-        if p and p not in phones:
-            phones.append(p)
-    name = str(contact.get("name") or "").strip()
-    if not name and emails:
-        name = emails[0].split("@")[0]
-    address = str(contact.get("address") or "").strip()
-    return {
-        "uid": str(contact.get("uid") or uuid.uuid4()),
-        "name": name,
-        "emails": emails,
-        "phones": phones,
-        "address": address,
-    }
-
-
-def _load_local_contacts() -> List[Dict]:
-    try:
-        if not LOCAL_CONTACTS_FILE.exists():
-            return []
-        data = json.loads(LOCAL_CONTACTS_FILE.read_text(encoding="utf-8"))
-        rows = data.get("contacts", data) if isinstance(data, dict) else data
-        return [_normalize_contact(c) for c in (rows or []) if isinstance(c, dict)]
-    except Exception as e:
-        logger.error(f"Failed to load local contacts: {e}")
-        return []
-
-
-def _save_local_contacts(contacts: List[Dict]) -> None:
-    from core.atomic_io import atomic_write_json
-    DATA_DIR.mkdir(parents=True, exist_ok=True)
-    atomic_write_json(str(LOCAL_CONTACTS_FILE), {"contacts": [_normalize_contact(c) for c in contacts]}, indent=2)
-    _contact_cache["contacts"] = [_normalize_contact(c) for c in contacts]
-    _contact_cache["fetched_at"] = datetime.utcnow()
-
-
-# ── vCard parsing ──
-
-def _vunesc(value: str) -> str:
-    """Reverse _vesc() — turn escaped vCard text back into the raw value.
-    Order matters: handle \\n/\\, /\\; first, backslash-unescape last."""
-    if not value:
-        return value
-    out = []
-    i = 0
-    while i < len(value):
-        ch = value[i]
-        if ch == "\\" and i + 1 < len(value):
-            nxt = value[i + 1]
-            if nxt in ("n", "N"):
-                out.append("\n")
-            elif nxt in (",", ";", "\\"):
-                out.append(nxt)
-            else:
-                out.append(nxt)
-            i += 2
-        else:
-            out.append(ch)
-            i += 1
-    return "".join(out)
-
-
-def _parse_vcards(text: str) -> List[Dict]:
-    """Parse a stream of vCards into dicts with name, email, phone."""
-    # Unfold RFC 6350 3.2 line folding first: a CRLF/LF followed by a single
-    # space or tab is a continuation of the previous logical line. Real
-    # CardDAV servers (Radicale, iCloud, Apple/Google) fold long EMAIL / FN /
-    # PHOTO lines, and splitting on raw newlines without unfolding dropped the
-    # continuation (e.g. "...@example\n .com" lost the ".com"), truncating the
-    # email/name.
-    text = re.sub(r"\r\n[ \t]", "", text or "")
-    text = re.sub(r"\n[ \t]", "", text)
-    contacts = []
-    for block in re.split(r"BEGIN:VCARD", text):
-        if not block.strip():
-            continue
-        contact = {"name": "", "emails": [], "phones": [], "uid": "", "address": ""}
-        for line in block.split("\n"):
-            line = line.strip()
-            # Strip an optional RFC 6350 group prefix (e.g. "item1.EMAIL;...")
-            # that Apple Contacts / iCloud / many CardDAV servers emit by
-            # default — without this the property-name checks below miss those
-            # lines and silently drop the email / phone. The group token only
-            # precedes the property name, so it is safe to strip for matching
-            # and value extraction, and a no-op for non-grouped lines.
-            name_part = re.sub(r"^[A-Za-z0-9-]+\.", "", line, count=1)
-            if name_part.startswith("FN:") or name_part.startswith("FN;"):
-                contact["name"] = _vunesc(name_part.split(":", 1)[1]) if ":" in name_part else ""
-            elif name_part.startswith("EMAIL"):
-                # Handle EMAIL:foo@bar OR EMAIL;TYPE=...:foo@bar OR EMAIL;PREF=1:foo@bar
-                if ":" in name_part:
-                    email_addr = _vunesc(name_part.split(":", 1)[1])
-                    if email_addr and email_addr not in contact["emails"]:
-                        contact["emails"].append(email_addr)
-            elif name_part.startswith("TEL"):
-                if ":" in name_part:
-                    phone = _vunesc(name_part.split(":", 1)[1])
-                    if phone and phone not in contact["phones"]:
-                        contact["phones"].append(phone)
-            elif name_part.startswith("ADR"):
-                # vCard ADR is 7 semicolon-separated components:
-                # post-office-box;extended-address;street;locality;region;postal-code;country.
-                # Recover a human-readable string by joining non-empty
-                # components with ", ".
-                if ":" in name_part:
-                    raw = name_part.split(":", 1)[1]
-                    parts = [_vunesc(p).strip() for p in raw.split(";")]
-                    contact["address"] = ", ".join(p for p in parts if p)
-            elif name_part.startswith("UID:"):
-                contact["uid"] = _vunesc(name_part[4:])
-        if contact["name"] or contact["emails"]:
-            contacts.append(contact)
-    return contacts
-
-
-def _vesc(value: str) -> str:
-    """Escape a vCard property VALUE per RFC 6350 §3.4: backslash, comma,
-    semicolon, and newlines. Without this, a name like 'Sekisui House,Ltd'
-    or any value containing a newline produces a malformed vCard (broken
-    N/FN fields) or could inject arbitrary properties."""
-    return (
-        (value or "")
-        .replace("\\", "\\\\")
-        .replace("\n", "\\n")
-        .replace("\r", "")
-        .replace(",", "\\,")
-        .replace(";", "\\;")
-    )
-
-
-def _build_vcard(name: str, email: str, uid: Optional[str] = None,
-                 emails: Optional[List[str]] = None,
-                 phones: Optional[List[str]] = None,
-                 address: Optional[str] = None) -> str:
-    """Build a vCard. Accepts either a single `email` (legacy callers) or
-    full `emails`/`phones` lists (edit path). The first email is marked
-    PREF=1. All values are RFC-6350-escaped."""
-    if not uid:
-        uid = str(uuid.uuid4())
-    # Normalize email lists — `email` arg is a convenience for single-email
-    # creation; `emails` (if given) is authoritative.
-    email_list = [e.strip() for e in (emails if emails is not None else ([email] if email else [])) if e and e.strip()]
-    phone_list = [p.strip() for p in (phones or []) if p and p.strip()]
-    # Try to split name into first/last
-    parts = name.strip().split()
-    if len(parts) >= 2:
-        first = parts[0]
-        last = " ".join(parts[1:])
-    else:
-        first = name
-        last = ""
-    # N field is structured (5 components separated by ';') — escape each
-    # component individually so a comma in the name doesn't split it.
-    n_field = f"{_vesc(last)};{_vesc(first)};;;"
-    lines = [
-        "BEGIN:VCARD",
-        "VERSION:4.0",
-        f"UID:{_vesc(uid)}",
-        f"FN:{_vesc(name)}",
-        f"N:{n_field}",
-    ]
-    for i, em in enumerate(email_list):
-        # First email is the preferred one.
-        lines.append(f"EMAIL;PREF=1:{_vesc(em)}" if i == 0 else f"EMAIL:{_vesc(em)}")
-    for ph in phone_list:
-        lines.append(f"TEL:{_vesc(ph)}")
-    # Address: stuff the whole human-readable string into the street
-    # component of ADR. vCard ADR has 7 semicolon-separated components:
-    # post-office-box;extended-address;street;locality;region;postal-code;country.
-    addr = (address or "").strip()
-    if addr:
-        lines.append(f"ADR:;;{_vesc(addr)};;;;")
-    lines.append("END:VCARD")
-    return "\r\n".join(lines) + "\r\n"
-
-
-# ── In-memory cache ──
-
-_contact_cache = {"contacts": [], "fetched_at": None}
-
-
-def _abs_url(href: str) -> str:
-    """Combine a multistatus <href> (an absolute path like
-    /user/contacts/x.vcf) with the configured CardDAV server origin so we
-    get a fully-qualified URL to PUT/DELETE. Absolute hrefs are accepted only
-    for the configured origin; a cross-origin href is treated as a path on the
-    configured server so a malicious CardDAV response cannot redirect later
-    writes/deletes to cloud metadata or another host."""
-    cfg = _get_carddav_config()
-    base = _carddav_base_url(cfg)
-    base_p = urlparse(base)
-    joined = urljoin(base.rstrip("/") + "/", href or "")
-    joined_p = urlparse(joined)
-    if (joined_p.scheme, joined_p.netloc) != (base_p.scheme, base_p.netloc):
-        joined = urlunparse((base_p.scheme, base_p.netloc, joined_p.path or "/", "", joined_p.query, ""))
-    return _validate_carddav_url(joined)
-
-
-# CardDAV REPORT body — pull every card's etag + raw vCard in ONE request,
-# alongside the resource href. Lets us map each contact's UID to the real
-# server resource path (which is NOT always <uid>.vcf for contacts created
-# by other clients).
-_ADDRESSBOOK_QUERY = (
-    '<?xml version="1.0" encoding="utf-8"?>'
-    '<C:addressbook-query xmlns:D="DAV:" xmlns:C="urn:ietf:params:xml:ns:carddav">'
-    '<D:prop><D:getetag/><C:address-data/></D:prop>'
-    '<C:filter/>'
-    '</C:addressbook-query>'
-)
-
-
-def _fetch_via_report(cfg, auth):
-    """Try a CardDAV REPORT addressbook-query — returns contacts WITH an
-    `href` field, or None if the server doesn't support it / errors."""
-    from defusedxml import ElementTree as ET
-    try:
-        r = httpx.request(
-            "REPORT", cfg["url"],
-            content=_ADDRESSBOOK_QUERY.encode("utf-8"),
-            headers={"Content-Type": "application/xml; charset=utf-8", "Depth": "1"},
-            auth=auth, timeout=10,
-        )
-        if r.status_code not in (207, 200):
-            return None
-        root = ET.fromstring(r.text)
-        ns = {"D": "DAV:", "C": "urn:ietf:params:xml:ns:carddav"}
-        out = []
-        for resp in root.findall("D:response", ns):
-            href_el = resp.find("D:href", ns)
-            data_el = resp.find(".//C:address-data", ns)
-            if href_el is None or data_el is None or not (data_el.text or "").strip():
-                continue
-            parsed = _parse_vcards(data_el.text)
-            if not parsed:
-                continue
-            c = parsed[0]
-            c["href"] = href_el.text.strip()
-            out.append(c)
-        # If the REPORT parsed to ZERO contacts, don't trust it — some
-        # CardDAV servers treat an empty <filter/> as "match nothing" and
-        # return a valid-but-empty 207. Return None so the caller falls
-        # back to the plain GET (which lists everything). A genuinely empty
-        # address book just costs one extra GET that also returns nothing.
-        if not out:
-            return None
-        return out
-    except Exception as e:
-        logger.warning(f"CardDAV REPORT failed, falling back to GET: {e}")
-        return None
-
-
-def _fetch_contacts(force=False):
-    """Fetch all contacts. Uses CardDAV when configured, otherwise local JSON."""
-    if not force and _contact_cache["fetched_at"]:
-        age = (datetime.utcnow() - _contact_cache["fetched_at"]).total_seconds()
-        if age < 60:
-            return _contact_cache["contacts"]
-
-    cfg = _get_carddav_config()
-    if not _carddav_configured(cfg):
-        contacts = _load_local_contacts()
-        _contact_cache["contacts"] = contacts
-        _contact_cache["fetched_at"] = datetime.utcnow()
-        return contacts
-
-    try:
-        cfg["url"] = _carddav_base_url(cfg)
-        auth = None
-        if cfg["username"]:
-            auth = (cfg["username"], cfg["password"])
-        # Preferred path: REPORT gives us hrefs for reliable edit/delete.
-        contacts = _fetch_via_report(cfg, auth)
-        if contacts is None:
-            # Fallback: plain GET, concatenated vCards, no hrefs.
-            r = httpx.get(cfg["url"], auth=auth, timeout=10)
-            if r.status_code != 200:
-                logger.warning(f"CardDAV returned {r.status_code}")
-                return _contact_cache["contacts"]
-            contacts = _parse_vcards(r.text)
-        _contact_cache["contacts"] = contacts
-        _contact_cache["fetched_at"] = datetime.utcnow()
-        return contacts
-    except Exception as e:
-        logger.error(f"Failed to fetch contacts: {e}")
-        return _contact_cache["contacts"]
-
-
-def _resolve_resource_url(uid: str) -> str:
-    """Map a contact UID to its real CardDAV resource URL. Uses the href
-    captured during fetch when available (handles contacts whose filename
-    != UID); falls back to the <uid>.vcf guess for app-created contacts or
-    when no href is known."""
-    def _lookup():
-        for c in _contact_cache.get("contacts", []):
-            if c.get("uid") == uid and c.get("href"):
-                return _abs_url(c["href"])
-        return None
-    found = _lookup()
-    if found:
-        return found
-    # Not in cache (or no href) — refresh once and retry before guessing.
-    try:
-        _fetch_contacts(force=True)
-    except Exception:
-        pass
-    return _lookup() or _vcard_url(uid)
-
-
-def _create_contact(name: str, email: str = "", address: str = "", phones: Optional[List[str]] = None) -> bool:
-    """Add a new contact via CardDAV or local contacts."""
-    email = (email or "").strip()
-    phone_list = [str(p or "").strip() for p in (phones or []) if str(p or "").strip()]
-    cfg = _get_carddav_config()
-    if not _carddav_configured(cfg):
-        contacts = _load_local_contacts()
-        email_l = email.lower()
-        for c in contacts:
-            if email_l and email_l in [e.lower() for e in c.get("emails", [])]:
-                return True
-            if phone_list and any(p in (c.get("phones") or []) for p in phone_list):
-                return True
-        contacts.append(_normalize_contact({
-            "name": name,
-            "emails": [email] if email else [],
-            "phones": phone_list,
-            "address": address,
-        }))
-        _save_local_contacts(contacts)
-        return True
-
-    contact_uid = str(uuid.uuid4())
-    vcard = _build_vcard(name, email, contact_uid, address=address, phones=phone_list)
-    try:
-        url = _carddav_base_url(cfg) + "/" + contact_uid + ".vcf"
-        auth = None
-        if cfg["username"]:
-            auth = (cfg["username"], cfg["password"])
-        r = httpx.put(
-            url,
-            data=vcard.encode("utf-8"),
-            headers={"Content-Type": "text/vcard; charset=utf-8"},
-            auth=auth,
-            timeout=10,
-        )
-        if r.status_code in (200, 201, 204):
-            # Invalidate cache
-            _contact_cache["fetched_at"] = None
-            return True
-        logger.warning(f"CardDAV PUT returned {r.status_code}: {r.text[:200]}")
-        return False
-    except Exception as e:
-        logger.error(f"Failed to create contact: {e}")
-        return False
-
-
-def _vcard_url(uid: str) -> str:
-    """The CardDAV resource URL for a given contact UID. The uid is URL-
-    encoded so a value containing '/', '..' or other path chars can't
-    escape the collection and target an arbitrary CardDAV resource."""
-    from urllib.parse import quote
-    cfg = _get_carddav_config()
-    return _carddav_base_url(cfg) + "/" + quote(uid, safe="") + ".vcf"
-
-
-def _import_vcards(text: str) -> Dict:
-    """Import a (possibly multi-card) .vcf blob. Each card is PUT to the
-    CardDAV server PRESERVING its full original content (ADR/ORG/photo/
-    etc.) — we don't rebuild it, just ensure it has VERSION + UID and
-    normalize line endings. Returns {imported, failed, total}."""
-    from urllib.parse import quote
-    cfg = _get_carddav_config()
-    if not cfg.get("url"):
-        parsed = _parse_vcards(text)
-        contacts = _load_local_contacts()
-        existing = {
-            e.lower()
-            for c in contacts
-            for e in (c.get("emails") or [])
-            if e
-        }
-        imported = 0
-        for c in parsed:
-            emails = [e for e in (c.get("emails") or []) if e]
-            if emails and any(e.lower() in existing for e in emails):
-                continue
-            contacts.append(_normalize_contact(c))
-            for e in emails:
-                existing.add(e.lower())
-            imported += 1
-        if imported:
-            _save_local_contacts(contacts)
-        return {"imported": imported, "failed": 0, "total": len(parsed)}
-    try:
-        base_url = _carddav_base_url(cfg)
-    except ValueError as e:
-        logger.warning("CardDAV import URL rejected: %s", e)
-        return {"imported": 0, "failed": 0, "total": 0, "error": str(e)}
-    auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
-    # Split into individual cards. re.split drops the BEGIN line, so we
-    # re-add it. Normalize CRLF.
-    raw = (text or "").replace("\r\n", "\n").replace("\r", "\n")
-    blocks = []
-    for chunk in raw.split("BEGIN:VCARD"):
-        chunk = chunk.strip()
-        if not chunk:
-            continue
-        # Trim anything after END:VCARD (defensive).
-        end = chunk.upper().find("END:VCARD")
-        body = chunk[: end + len("END:VCARD")] if end != -1 else chunk
-        blocks.append("BEGIN:VCARD\n" + body)
-    imported = 0
-    failed = 0
-    for block in blocks:
-        # Extract or assign a UID.
-        m = re.search(r"^UID:(.+)$", block, re.MULTILINE)
-        uid = (m.group(1).strip() if m else "") or str(uuid.uuid4())
-        if not m:
-            # Inject a UID right after the VERSION line (or after BEGIN).
-            if re.search(r"^VERSION:", block, re.MULTILINE):
-                block = re.sub(r"(^VERSION:.*$)", r"\1\nUID:" + uid, block, count=1, flags=re.MULTILINE)
-            else:
-                block = block.replace("BEGIN:VCARD", f"BEGIN:VCARD\nVERSION:4.0\nUID:{uid}", 1)
-        elif not re.search(r"^VERSION:", block, re.MULTILINE):
-            block = block.replace("BEGIN:VCARD", "BEGIN:VCARD\nVERSION:4.0", 1)
-        vcard = block.replace("\n", "\r\n") + "\r\n"
-        url = base_url + "/" + quote(uid, safe="") + ".vcf"
-        try:
-            r = httpx.put(
-                url, data=vcard.encode("utf-8"),
-                headers={"Content-Type": "text/vcard; charset=utf-8"},
-                auth=auth, timeout=15,
-            )
-            if r.status_code in (200, 201, 204):
-                imported += 1
-            else:
-                failed += 1
-                logger.warning(f"Import PUT {uid} returned {r.status_code}: {r.text[:120]}")
-        except Exception as e:
-            failed += 1
-            logger.error(f"Import PUT {uid} failed: {e}")
-    if imported:
-        _contact_cache["fetched_at"] = None
-    return {"imported": imported, "failed": failed, "total": len(blocks)}
-
-
-def _import_csv_contacts(text: str) -> Dict:
-    """Import contacts from CSV. Supports common headers:
-    name/full_name/display_name, email/email_address/e-mail, phone/tel.
-    Falls back to first columns as name,email,phone when no headers exist."""
-    raw = (text or "").strip()
-    if not raw:
-        return {"imported": 0, "failed": 0, "total": 0, "error": "No CSV data found"}
-
-    try:
-        sample = raw[:2048]
-        dialect = csv.Sniffer().sniff(sample)
-    except Exception:
-        dialect = csv.excel
-
-    stream = io.StringIO(raw)
-    try:
-        has_header = csv.Sniffer().has_header(raw[:2048])
-    except Exception:
-        has_header = True
-
-    rows = []
-    if has_header:
-        reader = csv.DictReader(stream, dialect=dialect)
-        for row in reader:
-            lowered = {str(k or "").strip().lower(): (v or "").strip() for k, v in row.items()}
-            name = (
-                lowered.get("name") or lowered.get("full name") or lowered.get("full_name")
-                or lowered.get("display name") or lowered.get("display_name")
-                or lowered.get("fn") or ""
-            )
-            email = (
-                lowered.get("email") or lowered.get("email address")
-                or lowered.get("email_address") or lowered.get("e-mail")
-                or lowered.get("mail") or ""
-            )
-            phone = lowered.get("phone") or lowered.get("telephone") or lowered.get("tel") or ""
-            rows.append((name, email, phone))
-    else:
-        stream.seek(0)
-        reader = csv.reader(stream, dialect=dialect)
-        for row in reader:
-            cols = [(c or "").strip() for c in row]
-            if not any(cols):
-                continue
-            rows.append((
-                cols[0] if len(cols) > 0 else "",
-                cols[1] if len(cols) > 1 else "",
-                cols[2] if len(cols) > 2 else "",
-            ))
-
-    imported = 0
-    failed = 0
-    total = 0
-    existing_emails = {
-        e.lower()
-        for c in _fetch_contacts()
-        for e in (c.get("emails") or [])
-        if e
-    }
-    for name, email, phone in rows:
-        email = (email or "").strip()
-        name = (name or "").strip() or (email.split("@")[0] if email else "")
-        if not email:
-            continue
-        total += 1
-        if email.lower() in existing_emails:
-            continue
-        ok = _create_contact(name, email)
-        if ok:
-            imported += 1
-            existing_emails.add(email.lower())
-            # If the CSV had a phone number, rewrite the just-created row
-            # through the richer update path so phone lands in CardDAV too.
-            if phone:
-                try:
-                    contacts = _fetch_contacts(force=True)
-                    created = next((c for c in contacts if email.lower() in [e.lower() for e in c.get("emails", [])]), None)
-                    if created and created.get("uid"):
-                        _update_contact(created["uid"], name, [email], [phone])
-                except Exception:
-                    pass
-        else:
-            failed += 1
-
-    if imported:
-        _contact_cache["fetched_at"] = None
-    return {"imported": imported, "failed": failed, "total": total}
-
-
-def _contacts_to_vcf(contacts: List[Dict]) -> str:
-    return "".join(
-        _build_vcard(
-            c.get("name") or ((c.get("emails") or [""])[0].split("@")[0] if c.get("emails") else "Contact"),
-            "",
-            uid=c.get("uid") or str(uuid.uuid4()),
-            emails=c.get("emails") or [],
-            phones=c.get("phones") or [],
-        )
-        for c in contacts
-    )
-
-
-def _contacts_to_csv(contacts: List[Dict]) -> str:
-    out = io.StringIO()
-    writer = csv.writer(out)
-    writer.writerow(["name", "email", "phone"])
-    for c in contacts:
-        emails = c.get("emails") or [""]
-        phones = c.get("phones") or [""]
-        max_len = max(len(emails), len(phones), 1)
-        for i in range(max_len):
-            writer.writerow([
-                c.get("name") or "",
-                emails[i] if i < len(emails) else "",
-                phones[i] if i < len(phones) else "",
-            ])
-    return out.getvalue()
-
-
-def _update_contact(uid: str, name: str, emails: List[str], phones: List[str], address: str = "") -> bool:
-    """Rewrite an existing contact via CardDAV or local contacts."""
-    cfg = _get_carddav_config()
-    if not _carddav_configured(cfg):
-        contacts = _load_local_contacts()
-        found = False
-        out = []
-        for c in contacts:
-            if c.get("uid") == uid:
-                # Preserve existing address when caller passes "" (only
-                # updating name/emails/phones, not touching address).
-                addr = address if address else c.get("address", "")
-                out.append(_normalize_contact({"uid": uid, "name": name, "emails": emails, "phones": phones, "address": addr}))
-                found = True
-            else:
-                out.append(c)
-        if not found:
-            out.append(_normalize_contact({"uid": uid, "name": name, "emails": emails, "phones": phones, "address": address}))
-        _save_local_contacts(out)
-        return True
-
-    vcard = _build_vcard(name, "", uid=uid, emails=emails, phones=phones, address=address)
-    # Use the real resource href (handles externally-created contacts whose
-    # filename != UID); falls back to the <uid>.vcf guess.
-    try:
-        url = _resolve_resource_url(uid)
-        auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
-        r = httpx.put(
-            url,
-            data=vcard.encode("utf-8"),
-            headers={"Content-Type": "text/vcard; charset=utf-8"},
-            auth=auth,
-            timeout=10,
-        )
-        if r.status_code in (200, 201, 204):
-            _contact_cache["fetched_at"] = None
-            return True
-        logger.warning(f"CardDAV update PUT returned {r.status_code}: {r.text[:200]}")
-        return False
-    except Exception as e:
-        logger.error(f"Failed to update contact: {e}")
-        return False
-
-
-def _delete_contact(uid: str) -> bool:
-    """Delete a contact via CardDAV or local contacts."""
-    cfg = _get_carddav_config()
-    if not _carddav_configured(cfg):
-        contacts = _load_local_contacts()
-        remaining = [c for c in contacts if c.get("uid") != uid]
-        _save_local_contacts(remaining)
-        return True
-
-    try:
-        url = _resolve_resource_url(uid)
-        auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
-        r = httpx.delete(url, auth=auth, timeout=10)
-        if r.status_code in (200, 204, 404):
-            # Invalidate cache so the next fetch sees the server truth.
-            _contact_cache["fetched_at"] = None
-            # Verify: force a fresh fetch and check the UID is actually gone.
-            # A 404 on the guessed URL ({uid}.vcf) can mean the contact
-            # lives at a different resource URL — the DELETE missed it but
-            # we'd silently report success. This check catches that.
-            fresh = _fetch_contacts(force=True)
-            still_there = any(c.get("uid") == uid for c in fresh)
-            if still_there:
-                logger.warning(
-                    f"CardDAV DELETE reported success for {uid} "
-                    f"but UID still present after re-fetch — "
-                    f"resource URL may differ from {redact_url(url)}"
-                )
-                return False
-            if r.status_code == 404:
-                logger.info(f"CardDAV DELETE 404 for {uid} — already gone")
-            return True
-        logger.warning(f"CardDAV DELETE returned {r.status_code}: {r.text[:200]}")
-        return False
-    except Exception as e:
-        logger.error(f"Failed to delete contact: {e}")
-        return False
-
-
-# ── Routes ──
-
-def setup_contacts_routes():
-    router = APIRouter(prefix="/api/contacts", tags=["contacts"])
-
-    @router.get("/list")
-    async def list_contacts(_admin: str = Depends(require_admin)):
-        """List all contacts."""
-        contacts = _fetch_contacts()
-        return {"contacts": contacts, "count": len(contacts)}
-
-    @router.get("/search")
-    async def search_contacts(q: str = Query(""), _admin: str = Depends(require_admin)):
-        """Search contacts by name or email. Returns up to 10 matches."""
-        contacts = _fetch_contacts()
-        if not q:
-            return {"results": []}
-        q_lower = q.lower()
-        results = []
-        for c in contacts:
-            if q_lower in c["name"].lower():
-                results.append(c)
-                continue
-            for em in c["emails"]:
-                if q_lower in em.lower():
-                    results.append(c)
-                    break
-        return {"results": results[:10]}
-
-    @router.post("/add")
-    async def add_contact(data: dict, _admin: str = Depends(require_admin)):
-        """Add a new contact."""
-        name = (data.get("name") or "").strip()
-        email = (data.get("email") or "").strip()
-        phone = (data.get("phone") or "").strip()
-        phones = [str(p or "").strip() for p in (data.get("phones") or []) if str(p or "").strip()]
-        if phone and phone not in phones:
-            phones.insert(0, phone)
-        address = (data.get("address") or "").strip()
-        if not name and email:
-            name = email.split("@")[0]
-        if not name and not email and not phones and not address:
-            return {"success": False, "error": "Name, email, phone, or address required"}
-        if not name:
-            name = email.split("@")[0] if email else (phones[0] if phones else "Contact")
-        contacts = _fetch_contacts()
-        for c in contacts:
-            if email and email.lower() in [e.lower() for e in c.get("emails", [])]:
-                return {"success": True, "message": "Already exists", "contact": c}
-            if phones and any(p in (c.get("phones") or []) for p in phones):
-                return {"success": True, "message": "Already exists", "contact": c}
-        create_params = inspect.signature(_create_contact).parameters
-        if "phones" in create_params:
-            ok = _create_contact(name, email, address, phones=phones)
-        elif len(create_params) >= 3:
-            ok = _create_contact(name, email, address)
-        else:
-            ok = _create_contact(name, email)
-        # If a phone was provided, do an immediate update to thread it
-        # through (the simple _create_contact signature only takes name +
-        # email + address; phones happen via update).
-        if ok and phones and "phones" not in create_params:
-            try:
-                fresh = _fetch_contacts(force=True)
-                created = next((c for c in fresh if name == c.get("name") and (not email or email in c.get("emails", []))), None)
-                if created:
-                    _update_contact(
-                        created["uid"], name,
-                        created.get("emails", []),
-                        phones,
-                        address,
-                    )
-            except Exception:
-                pass
-        return {"success": ok}
-
-    @router.post("/import")
-    async def import_vcf(data: dict, _admin: str = Depends(require_admin)):
-        """Import contacts from .vcf or CSV. Body: {"vcf": "..."} or {"csv": "..."}."""
-        # Coerce defensively: a non-string vcf/text/csv (e.g. a number or list
-        # in the JSON body) would otherwise reach .strip() and 500 with an
-        # AttributeError instead of degrading to a clean "no data" response.
-        text = str(data.get("vcf") or data.get("text") or "")
-        csv_text = str(data.get("csv") or "")
-        if text.strip():
-            if "BEGIN:VCARD" not in text.upper():
-                return {"success": False, "error": "No vCard data found"}
-            result = _import_vcards(text)
-        elif csv_text.strip():
-            result = _import_csv_contacts(csv_text)
-        else:
-            return {"success": False, "error": "No contact data found"}
-        result["success"] = result.get("imported", 0) > 0
-        return result
-
-    @router.get("/export")
-    async def export_contacts(
-        format: str = Query("vcf", pattern="^(vcf|csv)$"),
-        _admin: str = Depends(require_admin),
-    ):
-        """Export all contacts as vCard or CSV."""
-        contacts = _fetch_contacts(force=True)
-        if format == "csv":
-            content = _contacts_to_csv(contacts)
-            media_type = "text/csv; charset=utf-8"
-            filename = "odysseus-contacts.csv"
-        else:
-            content = _contacts_to_vcf(contacts)
-            media_type = "text/vcard; charset=utf-8"
-            filename = "odysseus-contacts.vcf"
-        return Response(
-            content=content,
-            media_type=media_type,
-            headers={"Content-Disposition": f'attachment; filename="{filename}"'},
-        )
-
-    @router.get("/config")
-    async def get_config(_admin: str = Depends(require_admin)):
-        cfg = _get_carddav_config()
-        # Mask password
-        if cfg["password"]:
-            cfg["password"] = "***"
-        return cfg
-
-    @router.put("/config")
-    async def update_config(data: dict, _admin: str = Depends(require_admin)):
-        settings = _load_settings()
-        for key in ("carddav_url", "carddav_username", "carddav_password"):
-            if key in data:
-                if key == "carddav_url" and str(data[key] or "").strip():
-                    try:
-                        settings[key] = _validate_carddav_url(data[key])
-                    except ValueError as e:
-                        raise HTTPException(400, str(e))
-                else:
-                    value = data[key]
-                    if key == "carddav_password" and value:
-                        from src.secret_storage import encrypt
-                        value = encrypt(value)
-                    settings[key] = value
-        _save_settings(settings)
-        # Force re-fetch
-        _contact_cache["fetched_at"] = None
-        return {"success": True}
-
-    @router.delete("/clear")
-    async def clear_contacts(_admin: str = Depends(require_admin)):
-        """Clear all local contacts. If CardDAV is configured, only clears the local fallback cache."""
-        _save_local_contacts([])
-        return {"success": True}
-
-    # NOTE: the /{uid} routes are declared LAST so the literal paths above
-    # (/list, /search, /add, /config) win — otherwise PUT /config would
-    # match PUT /{uid} with uid="config".
-    @router.put("/{uid}")
-    async def edit_contact(uid: str, data: dict, _admin: str = Depends(require_admin)):
-        """Edit an existing contact — name / emails / phones / address."""
-        name = (data.get("name") or "").strip()
-        emails = data.get("emails")
-        phones = data.get("phones")
-        if emails is None and data.get("email"):
-            emails = [data["email"]]
-        emails = [e.strip() for e in (emails or []) if e and e.strip()]
-        phones = [p.strip() for p in (phones or []) if p and p.strip()]
-        address = (data.get("address") or "").strip()
-        if not name and not emails and not address:
-            return {"success": False, "error": "Name, email, or address required"}
-        if not name and emails:
-            name = emails[0].split("@")[0]
-        ok = _update_contact(uid, name, emails, phones, address)
-        return {"success": ok}
-
-    @router.delete("/{uid}")
-    async def delete_contact(uid: str, _admin: str = Depends(require_admin)):
-        """Delete a contact by UID."""
-        if not uid:
-            return {"success": False, "error": "UID required"}
-        ok = _delete_contact(uid)
-        return {"success": ok}
-
-    return router

routes/contacts_routes.py

@@ -1,13 +1,892 @@
-"""Backward-compat shim — canonical location is routes/contacts/contacts_routes.py.
+"""
+contacts_routes.py
 
-This module is replaced in ``sys.modules`` by the canonical module object so
-that ``import routes.contacts_routes``, ``from routes.contacts_routes import X``,
-``importlib.import_module("routes.contacts_routes")``, and string-targeted
-monkeypatches all operate on the same object the application actually uses.
+CardDAV contacts integration. Reads from local Radicale, supports
+search and adding new contacts.
 """
 
-import sys as _sys
+import re
+import logging
+import uuid
+import json
+import csv
+import io
+import os
+import inspect
+import httpx
+from pathlib import Path
+from datetime import datetime
+from urllib.parse import urljoin, urlparse, urlunparse
 
-from routes.contacts import contacts_routes as _canonical  # noqa: F401
+from core.log_safety import redact_url
+from fastapi import APIRouter, Query, Depends, Response, HTTPException
+from typing import List, Dict, Optional
 
-_sys.modules[__name__] = _canonical
+from core.middleware import require_admin
+from src.url_safety import check_outbound_url
+
+logger = logging.getLogger(__name__)
+
+from src.constants import DATA_DIR as _DATA_DIR, SETTINGS_FILE as _SETTINGS_FILE, CONTACTS_FILE as _CONTACTS_FILE
+DATA_DIR = Path(_DATA_DIR)
+SETTINGS_FILE = Path(_SETTINGS_FILE)
+LOCAL_CONTACTS_FILE = Path(_CONTACTS_FILE)
+
+
+def _load_settings():
+    if SETTINGS_FILE.exists():
+        return json.loads(SETTINGS_FILE.read_text(encoding="utf-8"))
+    return {}
+
+
+def _save_settings(settings):
+    from core.atomic_io import atomic_write_json
+    atomic_write_json(str(SETTINGS_FILE), settings, indent=2)
+
+
+def _get_carddav_config():
+    import os
+    settings = _load_settings()
+    password = settings.get("carddav_password", os.environ.get("CARDDAV_PASSWORD", ""))
+    if password and "carddav_password" in settings:
+        from src.secret_storage import decrypt
+        password = decrypt(password)
+    return {
+        "url": settings.get("carddav_url", os.environ.get("CARDDAV_URL", "")),
+        "username": settings.get("carddav_username", os.environ.get("CARDDAV_USERNAME", "")),
+        "password": password,
+    }
+
+
+def _carddav_configured(cfg: Optional[Dict] = None) -> bool:
+    cfg = cfg or _get_carddav_config()
+    return bool((cfg.get("url") or "").strip())
+
+
+def _validate_carddav_url(url: str) -> str:
+    cleaned = (url if isinstance(url, str) else "").strip().rstrip("/")
+    ok, reason = check_outbound_url(
+        cleaned,
+        block_private=os.getenv("CARDDAV_BLOCK_PRIVATE_IPS", "false").lower() == "true",
+    )
+    if not ok:
+        raise ValueError(f"Rejected CardDAV URL: {reason}")
+    return cleaned
+
+
+def _carddav_base_url(cfg: Dict) -> str:
+    return _validate_carddav_url(cfg.get("url") or "")
+
+
+def _normalize_contact(contact: Dict) -> Dict:
+    emails = []
+    for e in contact.get("emails") or ([] if not contact.get("email") else [contact.get("email")]):
+        e = str(e or "").strip()
+        if e and e not in emails:
+            emails.append(e)
+    phones = []
+    for p in contact.get("phones") or ([] if not contact.get("phone") else [contact.get("phone")]):
+        p = str(p or "").strip()
+        if p and p not in phones:
+            phones.append(p)
+    name = str(contact.get("name") or "").strip()
+    if not name and emails:
+        name = emails[0].split("@")[0]
+    address = str(contact.get("address") or "").strip()
+    return {
+        "uid": str(contact.get("uid") or uuid.uuid4()),
+        "name": name,
+        "emails": emails,
+        "phones": phones,
+        "address": address,
+    }
+
+
+def _load_local_contacts() -> List[Dict]:
+    try:
+        if not LOCAL_CONTACTS_FILE.exists():
+            return []
+        data = json.loads(LOCAL_CONTACTS_FILE.read_text(encoding="utf-8"))
+        rows = data.get("contacts", data) if isinstance(data, dict) else data
+        return [_normalize_contact(c) for c in (rows or []) if isinstance(c, dict)]
+    except Exception as e:
+        logger.error(f"Failed to load local contacts: {e}")
+        return []
+
+
+def _save_local_contacts(contacts: List[Dict]) -> None:
+    from core.atomic_io import atomic_write_json
+    DATA_DIR.mkdir(parents=True, exist_ok=True)
+    atomic_write_json(str(LOCAL_CONTACTS_FILE), {"contacts": [_normalize_contact(c) for c in contacts]}, indent=2)
+    _contact_cache["contacts"] = [_normalize_contact(c) for c in contacts]
+    _contact_cache["fetched_at"] = datetime.utcnow()
+
+
+# ── vCard parsing ──
+
+def _vunesc(value: str) -> str:
+    """Reverse _vesc() — turn escaped vCard text back into the raw value.
+    Order matters: handle \\n/\\, /\\; first, backslash-unescape last."""
+    if not value:
+        return value
+    out = []
+    i = 0
+    while i < len(value):
+        ch = value[i]
+        if ch == "\\" and i + 1 < len(value):
+            nxt = value[i + 1]
+            if nxt in ("n", "N"):
+                out.append("\n")
+            elif nxt in (",", ";", "\\"):
+                out.append(nxt)
+            else:
+                out.append(nxt)
+            i += 2
+        else:
+            out.append(ch)
+            i += 1
+    return "".join(out)
+
+
+def _parse_vcards(text: str) -> List[Dict]:
+    """Parse a stream of vCards into dicts with name, email, phone."""
+    contacts = []
+    for block in re.split(r"BEGIN:VCARD", text):
+        if not block.strip():
+            continue
+        contact = {"name": "", "emails": [], "phones": [], "uid": "", "address": ""}
+        for line in block.split("\n"):
+            line = line.strip()
+            # Strip an optional RFC 6350 group prefix (e.g. "item1.EMAIL;...")
+            # that Apple Contacts / iCloud / many CardDAV servers emit by
+            # default — without this the property-name checks below miss those
+            # lines and silently drop the email / phone. The group token only
+            # precedes the property name, so it is safe to strip for matching
+            # and value extraction, and a no-op for non-grouped lines.
+            name_part = re.sub(r"^[A-Za-z0-9-]+\.", "", line, count=1)
+            if name_part.startswith("FN:") or name_part.startswith("FN;"):
+                contact["name"] = _vunesc(name_part.split(":", 1)[1]) if ":" in name_part else ""
+            elif name_part.startswith("EMAIL"):
+                # Handle EMAIL:foo@bar OR EMAIL;TYPE=...:foo@bar OR EMAIL;PREF=1:foo@bar
+                if ":" in name_part:
+                    email_addr = _vunesc(name_part.split(":", 1)[1])
+                    if email_addr and email_addr not in contact["emails"]:
+                        contact["emails"].append(email_addr)
+            elif name_part.startswith("TEL"):
+                if ":" in name_part:
+                    phone = _vunesc(name_part.split(":", 1)[1])
+                    if phone and phone not in contact["phones"]:
+                        contact["phones"].append(phone)
+            elif name_part.startswith("ADR"):
+                # vCard ADR is 7 semicolon-separated components:
+                # post-office-box;extended-address;street;locality;region;postal-code;country.
+                # Recover a human-readable string by joining non-empty
+                # components with ", ".
+                if ":" in name_part:
+                    raw = name_part.split(":", 1)[1]
+                    parts = [_vunesc(p).strip() for p in raw.split(";")]
+                    contact["address"] = ", ".join(p for p in parts if p)
+            elif name_part.startswith("UID:"):
+                contact["uid"] = _vunesc(name_part[4:])
+        if contact["name"] or contact["emails"]:
+            contacts.append(contact)
+    return contacts
+
+
+def _vesc(value: str) -> str:
+    """Escape a vCard property VALUE per RFC 6350 §3.4: backslash, comma,
+    semicolon, and newlines. Without this, a name like 'Sekisui House,Ltd'
+    or any value containing a newline produces a malformed vCard (broken
+    N/FN fields) or could inject arbitrary properties."""
+    return (
+        (value or "")
+        .replace("\\", "\\\\")
+        .replace("\n", "\\n")
+        .replace("\r", "")
+        .replace(",", "\\,")
+        .replace(";", "\\;")
+    )
+
+
+def _build_vcard(name: str, email: str, uid: Optional[str] = None,
+                 emails: Optional[List[str]] = None,
+                 phones: Optional[List[str]] = None,
+                 address: Optional[str] = None) -> str:
+    """Build a vCard. Accepts either a single `email` (legacy callers) or
+    full `emails`/`phones` lists (edit path). The first email is marked
+    PREF=1. All values are RFC-6350-escaped."""
+    if not uid:
+        uid = str(uuid.uuid4())
+    # Normalize email lists — `email` arg is a convenience for single-email
+    # creation; `emails` (if given) is authoritative.
+    email_list = [e.strip() for e in (emails if emails is not None else ([email] if email else [])) if e and e.strip()]
+    phone_list = [p.strip() for p in (phones or []) if p and p.strip()]
+    # Try to split name into first/last
+    parts = name.strip().split()
+    if len(parts) >= 2:
+        first = parts[0]
+        last = " ".join(parts[1:])
+    else:
+        first = name
+        last = ""
+    # N field is structured (5 components separated by ';') — escape each
+    # component individually so a comma in the name doesn't split it.
+    n_field = f"{_vesc(last)};{_vesc(first)};;;"
+    lines = [
+        "BEGIN:VCARD",
+        "VERSION:4.0",
+        f"UID:{_vesc(uid)}",
+        f"FN:{_vesc(name)}",
+        f"N:{n_field}",
+    ]
+    for i, em in enumerate(email_list):
+        # First email is the preferred one.
+        lines.append(f"EMAIL;PREF=1:{_vesc(em)}" if i == 0 else f"EMAIL:{_vesc(em)}")
+    for ph in phone_list:
+        lines.append(f"TEL:{_vesc(ph)}")
+    # Address: stuff the whole human-readable string into the street
+    # component of ADR. vCard ADR has 7 semicolon-separated components:
+    # post-office-box;extended-address;street;locality;region;postal-code;country.
+    addr = (address or "").strip()
+    if addr:
+        lines.append(f"ADR:;;{_vesc(addr)};;;;")
+    lines.append("END:VCARD")
+    return "\r\n".join(lines) + "\r\n"
+
+
+# ── In-memory cache ──
+
+_contact_cache = {"contacts": [], "fetched_at": None}
+
+
+def _abs_url(href: str) -> str:
+    """Combine a multistatus <href> (an absolute path like
+    /user/contacts/x.vcf) with the configured CardDAV server origin so we
+    get a fully-qualified URL to PUT/DELETE. Absolute hrefs are accepted only
+    for the configured origin; a cross-origin href is treated as a path on the
+    configured server so a malicious CardDAV response cannot redirect later
+    writes/deletes to cloud metadata or another host."""
+    cfg = _get_carddav_config()
+    base = _carddav_base_url(cfg)
+    base_p = urlparse(base)
+    joined = urljoin(base.rstrip("/") + "/", href or "")
+    joined_p = urlparse(joined)
+    if (joined_p.scheme, joined_p.netloc) != (base_p.scheme, base_p.netloc):
+        joined = urlunparse((base_p.scheme, base_p.netloc, joined_p.path or "/", "", joined_p.query, ""))
+    return _validate_carddav_url(joined)
+
+
+# CardDAV REPORT body — pull every card's etag + raw vCard in ONE request,
+# alongside the resource href. Lets us map each contact's UID to the real
+# server resource path (which is NOT always <uid>.vcf for contacts created
+# by other clients).
+_ADDRESSBOOK_QUERY = (
+    '<?xml version="1.0" encoding="utf-8"?>'
+    '<C:addressbook-query xmlns:D="DAV:" xmlns:C="urn:ietf:params:xml:ns:carddav">'
+    '<D:prop><D:getetag/><C:address-data/></D:prop>'
+    '<C:filter/>'
+    '</C:addressbook-query>'
+)
+
+
+def _fetch_via_report(cfg, auth):
+    """Try a CardDAV REPORT addressbook-query — returns contacts WITH an
+    `href` field, or None if the server doesn't support it / errors."""
+    from defusedxml import ElementTree as ET
+    try:
+        r = httpx.request(
+            "REPORT", cfg["url"],
+            content=_ADDRESSBOOK_QUERY.encode("utf-8"),
+            headers={"Content-Type": "application/xml; charset=utf-8", "Depth": "1"},
+            auth=auth, timeout=10,
+        )
+        if r.status_code not in (207, 200):
+            return None
+        root = ET.fromstring(r.text)
+        ns = {"D": "DAV:", "C": "urn:ietf:params:xml:ns:carddav"}
+        out = []
+        for resp in root.findall("D:response", ns):
+            href_el = resp.find("D:href", ns)
+            data_el = resp.find(".//C:address-data", ns)
+            if href_el is None or data_el is None or not (data_el.text or "").strip():
+                continue
+            parsed = _parse_vcards(data_el.text)
+            if not parsed:
+                continue
+            c = parsed[0]
+            c["href"] = href_el.text.strip()
+            out.append(c)
+        # If the REPORT parsed to ZERO contacts, don't trust it — some
+        # CardDAV servers treat an empty <filter/> as "match nothing" and
+        # return a valid-but-empty 207. Return None so the caller falls
+        # back to the plain GET (which lists everything). A genuinely empty
+        # address book just costs one extra GET that also returns nothing.
+        if not out:
+            return None
+        return out
+    except Exception as e:
+        logger.warning(f"CardDAV REPORT failed, falling back to GET: {e}")
+        return None
+
+
+def _fetch_contacts(force=False):
+    """Fetch all contacts. Uses CardDAV when configured, otherwise local JSON."""
+    if not force and _contact_cache["fetched_at"]:
+        age = (datetime.utcnow() - _contact_cache["fetched_at"]).total_seconds()
+        if age < 60:
+            return _contact_cache["contacts"]
+
+    cfg = _get_carddav_config()
+    if not _carddav_configured(cfg):
+        contacts = _load_local_contacts()
+        _contact_cache["contacts"] = contacts
+        _contact_cache["fetched_at"] = datetime.utcnow()
+        return contacts
+
+    try:
+        cfg["url"] = _carddav_base_url(cfg)
+        auth = None
+        if cfg["username"]:
+            auth = (cfg["username"], cfg["password"])
+        # Preferred path: REPORT gives us hrefs for reliable edit/delete.
+        contacts = _fetch_via_report(cfg, auth)
+        if contacts is None:
+            # Fallback: plain GET, concatenated vCards, no hrefs.
+            r = httpx.get(cfg["url"], auth=auth, timeout=10)
+            if r.status_code != 200:
+                logger.warning(f"CardDAV returned {r.status_code}")
+                return _contact_cache["contacts"]
+            contacts = _parse_vcards(r.text)
+        _contact_cache["contacts"] = contacts
+        _contact_cache["fetched_at"] = datetime.utcnow()
+        return contacts
+    except Exception as e:
+        logger.error(f"Failed to fetch contacts: {e}")
+        return _contact_cache["contacts"]
+
+
+def _resolve_resource_url(uid: str) -> str:
+    """Map a contact UID to its real CardDAV resource URL. Uses the href
+    captured during fetch when available (handles contacts whose filename
+    != UID); falls back to the <uid>.vcf guess for app-created contacts or
+    when no href is known."""
+    def _lookup():
+        for c in _contact_cache.get("contacts", []):
+            if c.get("uid") == uid and c.get("href"):
+                return _abs_url(c["href"])
+        return None
+    found = _lookup()
+    if found:
+        return found
+    # Not in cache (or no href) — refresh once and retry before guessing.
+    try:
+        _fetch_contacts(force=True)
+    except Exception:
+        pass
+    return _lookup() or _vcard_url(uid)
+
+
+def _create_contact(name: str, email: str, address: str = "") -> bool:
+    """Add a new contact via CardDAV or local contacts."""
+    cfg = _get_carddav_config()
+    if not _carddav_configured(cfg):
+        contacts = _load_local_contacts()
+        email_l = (email or "").strip().lower()
+        for c in contacts:
+            if email_l and email_l in [e.lower() for e in c.get("emails", [])]:
+                return True
+        contacts.append(_normalize_contact({"name": name, "emails": [email], "address": address}))
+        _save_local_contacts(contacts)
+        return True
+
+    contact_uid = str(uuid.uuid4())
+    vcard = _build_vcard(name, email, contact_uid, address=address)
+    try:
+        url = _carddav_base_url(cfg) + "/" + contact_uid + ".vcf"
+        auth = None
+        if cfg["username"]:
+            auth = (cfg["username"], cfg["password"])
+        r = httpx.put(
+            url,
+            data=vcard.encode("utf-8"),
+            headers={"Content-Type": "text/vcard; charset=utf-8"},
+            auth=auth,
+            timeout=10,
+        )
+        if r.status_code in (200, 201, 204):
+            # Invalidate cache
+            _contact_cache["fetched_at"] = None
+            return True
+        logger.warning(f"CardDAV PUT returned {r.status_code}: {r.text[:200]}")
+        return False
+    except Exception as e:
+        logger.error(f"Failed to create contact: {e}")
+        return False
+
+
+def _vcard_url(uid: str) -> str:
+    """The CardDAV resource URL for a given contact UID. The uid is URL-
+    encoded so a value containing '/', '..' or other path chars can't
+    escape the collection and target an arbitrary CardDAV resource."""
+    from urllib.parse import quote
+    cfg = _get_carddav_config()
+    return _carddav_base_url(cfg) + "/" + quote(uid, safe="") + ".vcf"
+
+
+def _import_vcards(text: str) -> Dict:
+    """Import a (possibly multi-card) .vcf blob. Each card is PUT to the
+    CardDAV server PRESERVING its full original content (ADR/ORG/photo/
+    etc.) — we don't rebuild it, just ensure it has VERSION + UID and
+    normalize line endings. Returns {imported, failed, total}."""
+    from urllib.parse import quote
+    cfg = _get_carddav_config()
+    if not cfg.get("url"):
+        parsed = _parse_vcards(text)
+        contacts = _load_local_contacts()
+        existing = {
+            e.lower()
+            for c in contacts
+            for e in (c.get("emails") or [])
+            if e
+        }
+        imported = 0
+        for c in parsed:
+            emails = [e for e in (c.get("emails") or []) if e]
+            if emails and any(e.lower() in existing for e in emails):
+                continue
+            contacts.append(_normalize_contact(c))
+            for e in emails:
+                existing.add(e.lower())
+            imported += 1
+        if imported:
+            _save_local_contacts(contacts)
+        return {"imported": imported, "failed": 0, "total": len(parsed)}
+    try:
+        base_url = _carddav_base_url(cfg)
+    except ValueError as e:
+        logger.warning("CardDAV import URL rejected: %s", e)
+        return {"imported": 0, "failed": 0, "total": 0, "error": str(e)}
+    auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
+    # Split into individual cards. re.split drops the BEGIN line, so we
+    # re-add it. Normalize CRLF.
+    raw = (text or "").replace("\r\n", "\n").replace("\r", "\n")
+    blocks = []
+    for chunk in raw.split("BEGIN:VCARD"):
+        chunk = chunk.strip()
+        if not chunk:
+            continue
+        # Trim anything after END:VCARD (defensive).
+        end = chunk.upper().find("END:VCARD")
+        body = chunk[: end + len("END:VCARD")] if end != -1 else chunk
+        blocks.append("BEGIN:VCARD\n" + body)
+    imported = 0
+    failed = 0
+    for block in blocks:
+        # Extract or assign a UID.
+        m = re.search(r"^UID:(.+)$", block, re.MULTILINE)
+        uid = (m.group(1).strip() if m else "") or str(uuid.uuid4())
+        if not m:
+            # Inject a UID right after the VERSION line (or after BEGIN).
+            if re.search(r"^VERSION:", block, re.MULTILINE):
+                block = re.sub(r"(^VERSION:.*$)", r"\1\nUID:" + uid, block, count=1, flags=re.MULTILINE)
+            else:
+                block = block.replace("BEGIN:VCARD", f"BEGIN:VCARD\nVERSION:4.0\nUID:{uid}", 1)
+        elif not re.search(r"^VERSION:", block, re.MULTILINE):
+            block = block.replace("BEGIN:VCARD", "BEGIN:VCARD\nVERSION:4.0", 1)
+        vcard = block.replace("\n", "\r\n") + "\r\n"
+        url = base_url + "/" + quote(uid, safe="") + ".vcf"
+        try:
+            r = httpx.put(
+                url, data=vcard.encode("utf-8"),
+                headers={"Content-Type": "text/vcard; charset=utf-8"},
+                auth=auth, timeout=15,
+            )
+            if r.status_code in (200, 201, 204):
+                imported += 1
+            else:
+                failed += 1
+                logger.warning(f"Import PUT {uid} returned {r.status_code}: {r.text[:120]}")
+        except Exception as e:
+            failed += 1
+            logger.error(f"Import PUT {uid} failed: {e}")
+    if imported:
+        _contact_cache["fetched_at"] = None
+    return {"imported": imported, "failed": failed, "total": len(blocks)}
+
+
+def _import_csv_contacts(text: str) -> Dict:
+    """Import contacts from CSV. Supports common headers:
+    name/full_name/display_name, email/email_address/e-mail, phone/tel.
+    Falls back to first columns as name,email,phone when no headers exist."""
+    raw = (text or "").strip()
+    if not raw:
+        return {"imported": 0, "failed": 0, "total": 0, "error": "No CSV data found"}
+
+    try:
+        sample = raw[:2048]
+        dialect = csv.Sniffer().sniff(sample)
+    except Exception:
+        dialect = csv.excel
+
+    stream = io.StringIO(raw)
+    try:
+        has_header = csv.Sniffer().has_header(raw[:2048])
+    except Exception:
+        has_header = True
+
+    rows = []
+    if has_header:
+        reader = csv.DictReader(stream, dialect=dialect)
+        for row in reader:
+            lowered = {str(k or "").strip().lower(): (v or "").strip() for k, v in row.items()}
+            name = (
+                lowered.get("name") or lowered.get("full name") or lowered.get("full_name")
+                or lowered.get("display name") or lowered.get("display_name")
+                or lowered.get("fn") or ""
+            )
+            email = (
+                lowered.get("email") or lowered.get("email address")
+                or lowered.get("email_address") or lowered.get("e-mail")
+                or lowered.get("mail") or ""
+            )
+            phone = lowered.get("phone") or lowered.get("telephone") or lowered.get("tel") or ""
+            rows.append((name, email, phone))
+    else:
+        stream.seek(0)
+        reader = csv.reader(stream, dialect=dialect)
+        for row in reader:
+            cols = [(c or "").strip() for c in row]
+            if not any(cols):
+                continue
+            rows.append((
+                cols[0] if len(cols) > 0 else "",
+                cols[1] if len(cols) > 1 else "",
+                cols[2] if len(cols) > 2 else "",
+            ))
+
+    imported = 0
+    failed = 0
+    total = 0
+    existing_emails = {
+        e.lower()
+        for c in _fetch_contacts()
+        for e in (c.get("emails") or [])
+        if e
+    }
+    for name, email, phone in rows:
+        email = (email or "").strip()
+        name = (name or "").strip() or (email.split("@")[0] if email else "")
+        if not email:
+            continue
+        total += 1
+        if email.lower() in existing_emails:
+            continue
+        ok = _create_contact(name, email)
+        if ok:
+            imported += 1
+            existing_emails.add(email.lower())
+            # If the CSV had a phone number, rewrite the just-created row
+            # through the richer update path so phone lands in CardDAV too.
+            if phone:
+                try:
+                    contacts = _fetch_contacts(force=True)
+                    created = next((c for c in contacts if email.lower() in [e.lower() for e in c.get("emails", [])]), None)
+                    if created and created.get("uid"):
+                        _update_contact(created["uid"], name, [email], [phone])
+                except Exception:
+                    pass
+        else:
+            failed += 1
+
+    if imported:
+        _contact_cache["fetched_at"] = None
+    return {"imported": imported, "failed": failed, "total": total}
+
+
+def _contacts_to_vcf(contacts: List[Dict]) -> str:
+    return "".join(
+        _build_vcard(
+            c.get("name") or ((c.get("emails") or [""])[0].split("@")[0] if c.get("emails") else "Contact"),
+            "",
+            uid=c.get("uid") or str(uuid.uuid4()),
+            emails=c.get("emails") or [],
+            phones=c.get("phones") or [],
+        )
+        for c in contacts
+    )
+
+
+def _contacts_to_csv(contacts: List[Dict]) -> str:
+    out = io.StringIO()
+    writer = csv.writer(out)
+    writer.writerow(["name", "email", "phone"])
+    for c in contacts:
+        emails = c.get("emails") or [""]
+        phones = c.get("phones") or [""]
+        max_len = max(len(emails), len(phones), 1)
+        for i in range(max_len):
+            writer.writerow([
+                c.get("name") or "",
+                emails[i] if i < len(emails) else "",
+                phones[i] if i < len(phones) else "",
+            ])
+    return out.getvalue()
+
+
+def _update_contact(uid: str, name: str, emails: List[str], phones: List[str], address: str = "") -> bool:
+    """Rewrite an existing contact via CardDAV or local contacts."""
+    cfg = _get_carddav_config()
+    if not _carddav_configured(cfg):
+        contacts = _load_local_contacts()
+        found = False
+        out = []
+        for c in contacts:
+            if c.get("uid") == uid:
+                # Preserve existing address when caller passes "" (only
+                # updating name/emails/phones, not touching address).
+                addr = address if address else c.get("address", "")
+                out.append(_normalize_contact({"uid": uid, "name": name, "emails": emails, "phones": phones, "address": addr}))
+                found = True
+            else:
+                out.append(c)
+        if not found:
+            out.append(_normalize_contact({"uid": uid, "name": name, "emails": emails, "phones": phones, "address": address}))
+        _save_local_contacts(out)
+        return True
+
+    vcard = _build_vcard(name, "", uid=uid, emails=emails, phones=phones, address=address)
+    # Use the real resource href (handles externally-created contacts whose
+    # filename != UID); falls back to the <uid>.vcf guess.
+    try:
+        url = _resolve_resource_url(uid)
+        auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
+        r = httpx.put(
+            url,
+            data=vcard.encode("utf-8"),
+            headers={"Content-Type": "text/vcard; charset=utf-8"},
+            auth=auth,
+            timeout=10,
+        )
+        if r.status_code in (200, 201, 204):
+            _contact_cache["fetched_at"] = None
+            return True
+        logger.warning(f"CardDAV update PUT returned {r.status_code}: {r.text[:200]}")
+        return False
+    except Exception as e:
+        logger.error(f"Failed to update contact: {e}")
+        return False
+
+
+def _delete_contact(uid: str) -> bool:
+    """Delete a contact via CardDAV or local contacts."""
+    cfg = _get_carddav_config()
+    if not _carddav_configured(cfg):
+        contacts = _load_local_contacts()
+        remaining = [c for c in contacts if c.get("uid") != uid]
+        _save_local_contacts(remaining)
+        return True
+
+    try:
+        url = _resolve_resource_url(uid)
+        auth = (cfg["username"], cfg["password"]) if cfg["username"] else None
+        r = httpx.delete(url, auth=auth, timeout=10)
+        if r.status_code in (200, 204, 404):
+            # Invalidate cache so the next fetch sees the server truth.
+            _contact_cache["fetched_at"] = None
+            # Verify: force a fresh fetch and check the UID is actually gone.
+            # A 404 on the guessed URL ({uid}.vcf) can mean the contact
+            # lives at a different resource URL — the DELETE missed it but
+            # we'd silently report success. This check catches that.
+            fresh = _fetch_contacts(force=True)
+            still_there = any(c.get("uid") == uid for c in fresh)
+            if still_there:
+                logger.warning(
+                    f"CardDAV DELETE reported success for {uid} "
+                    f"but UID still present after re-fetch — "
+                    f"resource URL may differ from {redact_url(url)}"
+                )
+                return False
+            if r.status_code == 404:
+                logger.info(f"CardDAV DELETE 404 for {uid} — already gone")
+            return True
+        logger.warning(f"CardDAV DELETE returned {r.status_code}: {r.text[:200]}")
+        return False
+    except Exception as e:
+        logger.error(f"Failed to delete contact: {e}")
+        return False
+
+
+# ── Routes ──
+
+def setup_contacts_routes():
+    router = APIRouter(prefix="/api/contacts", tags=["contacts"])
+
+    @router.get("/list")
+    async def list_contacts(_admin: str = Depends(require_admin)):
+        """List all contacts."""
+        contacts = _fetch_contacts()
+        return {"contacts": contacts, "count": len(contacts)}
+
+    @router.get("/search")
+    async def search_contacts(q: str = Query(""), _admin: str = Depends(require_admin)):
+        """Search contacts by name or email. Returns up to 10 matches."""
+        contacts = _fetch_contacts()
+        if not q:
+            return {"results": []}
+        q_lower = q.lower()
+        results = []
+        for c in contacts:
+            if q_lower in c["name"].lower():
+                results.append(c)
+                continue
+            for em in c["emails"]:
+                if q_lower in em.lower():
+                    results.append(c)
+                    break
+        return {"results": results[:10]}
+
+    @router.post("/add")
+    async def add_contact(data: dict, _admin: str = Depends(require_admin)):
+        """Add a new contact."""
+        name = (data.get("name") or "").strip()
+        email = (data.get("email") or "").strip()
+        phone = (data.get("phone") or "").strip()
+        address = (data.get("address") or "").strip()
+        if not email:
+            return {"success": False, "error": "Email required"}
+        # Check if already exists by email
+        if email:
+            contacts = _fetch_contacts()
+            for c in contacts:
+                if email.lower() in [e.lower() for e in c["emails"]]:
+                    return {"success": True, "message": "Already exists", "contact": c}
+        if not name:
+            name = email.split("@")[0]
+        create_params = inspect.signature(_create_contact).parameters
+        if len(create_params) >= 3:
+            ok = _create_contact(name, email, address)
+        else:
+            ok = _create_contact(name, email)
+        # If a phone was provided, do an immediate update to thread it
+        # through (the simple _create_contact signature only takes name +
+        # email + address; phones happen via update).
+        if ok and phone:
+            try:
+                fresh = _fetch_contacts(force=True)
+                created = next((c for c in fresh if name == c.get("name") and (not email or email in c.get("emails", []))), None)
+                if created:
+                    _update_contact(
+                        created["uid"], name,
+                        created.get("emails", []),
+                        [phone],
+                        address,
+                    )
+            except Exception:
+                pass
+        return {"success": ok}
+
+    @router.post("/import")
+    async def import_vcf(data: dict, _admin: str = Depends(require_admin)):
+        """Import contacts from .vcf or CSV. Body: {"vcf": "..."} or {"csv": "..."}."""
+        # Coerce defensively: a non-string vcf/text/csv (e.g. a number or list
+        # in the JSON body) would otherwise reach .strip() and 500 with an
+        # AttributeError instead of degrading to a clean "no data" response.
+        text = str(data.get("vcf") or data.get("text") or "")
+        csv_text = str(data.get("csv") or "")
+        if text.strip():
+            if "BEGIN:VCARD" not in text.upper():
+                return {"success": False, "error": "No vCard data found"}
+            result = _import_vcards(text)
+        elif csv_text.strip():
+            result = _import_csv_contacts(csv_text)
+        else:
+            return {"success": False, "error": "No contact data found"}
+        result["success"] = result.get("imported", 0) > 0
+        return result
+
+    @router.get("/export")
+    async def export_contacts(
+        format: str = Query("vcf", pattern="^(vcf|csv)$"),
+        _admin: str = Depends(require_admin),
+    ):
+        """Export all contacts as vCard or CSV."""
+        contacts = _fetch_contacts(force=True)
+        if format == "csv":
+            content = _contacts_to_csv(contacts)
+            media_type = "text/csv; charset=utf-8"
+            filename = "odysseus-contacts.csv"
+        else:
+            content = _contacts_to_vcf(contacts)
+            media_type = "text/vcard; charset=utf-8"
+            filename = "odysseus-contacts.vcf"
+        return Response(
+            content=content,
+            media_type=media_type,
+            headers={"Content-Disposition": f'attachment; filename="{filename}"'},
+        )
+
+    @router.get("/config")
+    async def get_config(_admin: str = Depends(require_admin)):
+        cfg = _get_carddav_config()
+        # Mask password
+        if cfg["password"]:
+            cfg["password"] = "***"
+        return cfg
+
+    @router.put("/config")
+    async def update_config(data: dict, _admin: str = Depends(require_admin)):
+        settings = _load_settings()
+        for key in ("carddav_url", "carddav_username", "carddav_password"):
+            if key in data:
+                if key == "carddav_url" and str(data[key] or "").strip():
+                    try:
+                        settings[key] = _validate_carddav_url(data[key])
+                    except ValueError as e:
+                        raise HTTPException(400, str(e))
+                else:
+                    value = data[key]
+                    if key == "carddav_password" and value:
+                        from src.secret_storage import encrypt
+                        value = encrypt(value)
+                    settings[key] = value
+        _save_settings(settings)
+        # Force re-fetch
+        _contact_cache["fetched_at"] = None
+        return {"success": True}
+
+    @router.delete("/clear")
+    async def clear_contacts(_admin: str = Depends(require_admin)):
+        """Clear all local contacts. If CardDAV is configured, only clears the local fallback cache."""
+        _save_local_contacts([])
+        return {"success": True}
+
+    # NOTE: the /{uid} routes are declared LAST so the literal paths above
+    # (/list, /search, /add, /config) win — otherwise PUT /config would
+    # match PUT /{uid} with uid="config".
+    @router.put("/{uid}")
+    async def edit_contact(uid: str, data: dict, _admin: str = Depends(require_admin)):
+        """Edit an existing contact — name / emails / phones / address."""
+        name = (data.get("name") or "").strip()
+        emails = data.get("emails")
+        phones = data.get("phones")
+        if emails is None and data.get("email"):
+            emails = [data["email"]]
+        emails = [e.strip() for e in (emails or []) if e and e.strip()]
+        phones = [p.strip() for p in (phones or []) if p and p.strip()]
+        address = (data.get("address") or "").strip()
+        if not name and not emails and not address:
+            return {"success": False, "error": "Name, email, or address required"}
+        if not name and emails:
+            name = emails[0].split("@")[0]
+        ok = _update_contact(uid, name, emails, phones, address)
+        return {"success": ok}
+
+    @router.delete("/{uid}")
+    async def delete_contact(uid: str, _admin: str = Depends(require_admin)):
+        """Delete a contact by UID."""
+        if not uid:
+            return {"success": False, "error": "UID required"}
+        ok = _delete_contact(uid)
+        return {"success": ok}
+
+    return router

routes/cookbook_helpers.py

@@ -439,30 +439,15 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
         "                if f.is_file(): nf += 1; sz += f.stat().st_size",
         "                if f.name.endswith('.incomplete'): ic = True",
         "        snap = os.path.join(cache, d, 'snapshots')",
-        "        def snapshot_size():",
-        "            total, count, incomplete = 0, 0, False",
-        "            seen_real = set()",
+        "        # Windows HF cache stores files directly in snapshots/; blobs/ may be empty.",
+        "        # Fallback: scan snapshots for real files when blobs yielded nothing.",
+        "        if sz == 0 and os.path.isdir(snap):",
         "            for sd in os.listdir(snap):",
         "                sf = os.path.join(snap, sd)",
         "                if not os.path.isdir(sf): continue",
-        "                for root, dirs, fns in safe_walk(sf):",
-        "                    for fn in fns:",
-        "                        fp = os.path.join(root, fn)",
-        "                        if fn.endswith('.incomplete'): incomplete = True",
-        "                        try:",
-        "                            real = os.path.realpath(fp)",
-        "                            if real in seen_real: continue",
-        "                            seen_real.add(real)",
-        "                            total += os.path.getsize(real)",
-        "                            count += 1",
-        "                        except Exception:",
-        "                            pass",
-        "            return total, count, incomplete",
-        "        # Some HF caches (macOS/MLX/Xet-style) keep blobs elsewhere or expose",
-        "        # snapshot symlinks only. Size snapshots too when blob accounting is empty.",
-        "        if sz == 0 and os.path.isdir(snap):",
-        "            sz2, nf2, ic2 = snapshot_size()",
-        "            sz, nf, ic = sz2, nf2, ic or ic2",
+        "                for f in os.scandir(sf):",
+        "                    if f.is_file(): nf += 1; sz += f.stat().st_size",
+        "                    if f.name.endswith('.incomplete'): ic = True",
         "        is_diffusion = False; gguf_files = []",
         "        if os.path.isdir(snap):",
         "            for sd in os.listdir(snap):",
@@ -486,18 +471,7 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
         "    add('/app/.cache/huggingface/hub')",
         f"    add({add_hf_cache!r})" if add_hf_cache else "",
         "    return candidates",
-        "def normalize_model_dir(p):",
-        "    p = os.path.expanduser((p or '').strip())",
-        "    if not p: return p",
-        "    if os.path.isdir(p) or os.path.isabs(p): return p",
-        "    # Users often paste Linux absolute paths without the leading slash.",
-        "    # Treat home/<user>/... as /home/<user>/... so remote scans work.",
-        "    if p.startswith(('home/', 'mnt/', 'media/', 'data/', 'opt/', 'srv/', 'var/')):",
-        "        prefixed = '/' + p",
-        "        if os.path.isdir(prefixed): return prefixed",
-        "    return p",
         "def scan_dir(p):",
-        "    p = normalize_model_dir(p)",
         "    if not os.path.isdir(p) or not safe_path(p): return",
         "    for d in sorted(os.listdir(p)):",
         "        if d.startswith('.'): continue",
@@ -567,7 +541,7 @@ def _cached_model_scan_script(model_dirs: list[str] | None = None, add_hf_cache:
         "scan_ollama()",
     ]
     for model_dir in model_dirs or []:
-        lines.append(f"scan_dir({model_dir!r})")
+        lines.append(f"scan_dir(os.path.expanduser({model_dir!r}))")
     lines.append("print(json.dumps(models))")
     return "\n".join(lines) + "\n"
 
@@ -584,22 +558,10 @@ def _bash_squote(v: str) -> str:
     return v.replace("'", "'\\''")
 
 
-# Shown by generated runner scripts when the ollama binary is missing on the
-# target host. Must stay free of backticks/$( ) and be emitted single-quoted:
-# an earlier version wrapped the install one-liner in backticks inside a
-# double-quoted echo, which bash executed as command substitution and ran the
-# system-wide installer (including on remote SSH hosts) instead of printing
-# the hint.
-OLLAMA_MISSING_HINT = (
-    "ERROR: Ollama not found on this server. Install it from "
-    "https://ollama.com/download or run: curl -fsSL https://ollama.com/install.sh | sh"
-)
-
-
 # Allow-list of binaries permitted as the leading token of `req.cmd` for /api/model/serve.
 # Anything else is rejected before the cmd is interpolated into a tmux/PowerShell wrapper.
 _SERVE_CMD_ALLOWLIST = {
-    "vllm", "llama-server", "llama-server.exe", "llama_server", "llama.cpp", "ollama",
+    "vllm", "llama-server", "llama_server", "llama.cpp", "ollama",
     "python", "python3",
     "sglang", "lmdeploy",
     "node", "npx",
@@ -615,16 +577,6 @@ _SERVE_CMD_ALLOWLIST = {
 _GGUF_PRELUDE_RE = re.compile(
     r'^MODEL_FILE=\$\([^\n]*?\)\s*&&\s*\{[^{}]*\}\s*\|\|\s*\{[^{}]*\}\s*&&\s*'
 )
-_SAFE_SUBSHELL_TEXT = r"[^'\n;&|`$()<>]+"
-_SAFE_SUBSHELL_DQ_HOME_PATH = r'"\$HOME/[^"\n;&|`()<>]*"'
-_SAFE_PRINTF_SUBSHELL_RE = re.compile(
-    rf"^\$\(printf[ \t]+%s[ \t]+(?:'{_SAFE_SUBSHELL_TEXT}'|\$\{{HOME\}}'/{_SAFE_SUBSHELL_TEXT}')\)$"
-)
-_SAFE_FIND_MMPROJ_SUBSHELL_RE = re.compile(
-    rf"^\$\(find[ \t]+(?:'{_SAFE_SUBSHELL_TEXT}'|{_SAFE_SUBSHELL_DQ_HOME_PATH}|{_SAFE_SUBSHELL_TEXT})"
-    r"[ \t]+-iname[ \t]+'mmproj\*\.gguf'"
-    r"(?:[ \t]+2>/dev/null)?[ \t]*\|[ \t]*sort[ \t]*\|[ \t]*head[ \t]+-1\)$"
-)
 _OLLAMA_HOST_ASSIGNMENT_RE = re.compile(r"(?:^|\s)OLLAMA_HOST=([^\s]+)")
 _OLLAMA_BIND_RE = re.compile(r"^\[([^\]]+)\]:(\d+)$|^([^:]+):(\d+)$")
 _OLLAMA_BIND_HOST_RE = re.compile(r"^[A-Za-z0-9._:-]+$")
@@ -725,13 +677,6 @@ def _check_serve_binary(seg: str) -> None:
         )
 
 
-def _is_safe_serve_subshell(subshell: str) -> bool:
-    return bool(
-        _SAFE_PRINTF_SUBSHELL_RE.fullmatch(subshell)
-        or _SAFE_FIND_MMPROJ_SUBSHELL_RE.fullmatch(subshell)
-    )
-
-
 def _validate_serve_cmd(v: str | None) -> str | None:
     """Reject serve commands that aren't in the allowlist or contain shell metachars.
 
@@ -763,15 +708,15 @@ def _validate_serve_cmd(v: str | None) -> str | None:
             _check_serve_binary(part.strip())
         return v
 
-    # Otherwise: a single invocation — no shell metacharacters allowed. Replace
-    # only the exact command substitutions emitted by the Cookbook UI:
-    # $(printf %s 'safe-path') and the mmproj lookup
-    # $(find <path> -iname 'mmproj*.gguf' 2>/dev/null | sort | head -1).
-    def _replace_safe_subshell(match: re.Match[str]) -> str:
-        subshell = match.group(0)
-        return "/placeholder/safe/path" if _is_safe_serve_subshell(subshell) else subshell
-
-    cleaned_v = re.sub(r"\$\([^()]*\)", _replace_safe_subshell, v)
+    # Otherwise: a single invocation — no shell metacharacters allowed.
+    # Temporarily replace safe $(printf %s ...) expressions with a placeholder
+    # to avoid triggering the metacharacter/command-injection checks.
+    cleaned_v = v
+    printf_matches = list(re.finditer(r"\$\(\s*printf\s+%s\s+([^\n()]*?)\)", v))
+    for match in printf_matches:
+        inner = match.group(1)
+        if not any(c in inner for c in (";", "&&", "||", "$(", "`")):
+            cleaned_v = cleaned_v.replace(match.group(0), "/placeholder/safe/path.gguf")
 
     # (`$(` was the original intent; bare `$` is fine for shell-safe paths.)
     if any(c in cleaned_v for c in (";", "&&", "||", "$(")):
@@ -1299,15 +1244,13 @@ def _diagnose_serve_output(text: str) -> dict | None:
             [{"label": "install vLLM in Cookbook Dependencies", "op": "dependency", "package": "vllm"}],
         ),
         (
-            r"sgl_kernel[\s\S]*(Python\.h|libnuma\.so\.1|common_ops|libnvrtc\.so)|"
-            r"(Python\.h|libnuma\.so\.1|common_ops|libnvrtc\.so)[\s\S]*sgl_kernel|"
-            r"Could not load any common_ops library|"
+            r"sgl_kernel[\s\S]*(Python\.h|libnuma\.so\.1|common_ops)|"
+            r"(Python\.h|libnuma\.so\.1|common_ops)[\s\S]*sgl_kernel|"
             r"Please ensure sgl_kernel is properly installed",
-            "SGLang native kernel/runtime is missing or mismatched on this server.",
+            "SGLang native dependencies are missing on this server.",
             [
-                {"label": "repair sglang-kernel in this Python environment", "op": "dependency", "package": "sglang-kernel"},
                 {"label": "install OS packages: libnuma-dev python3.12-dev build-essential", "op": "manual"},
-                {"label": "if libnvrtc is still missing, install the matching CUDA/NVRTC runtime on this host", "op": "manual"},
+                {"label": "upgrade sglang-kernel after OS packages are installed", "op": "manual"},
             ],
         ),
         (
@@ -1315,19 +1258,6 @@ def _diagnose_serve_output(text: str) -> dict | None:
             "SGLang is not installed or not in PATH on this server.",
             [{"label": "install SGLang in Cookbook Dependencies", "op": "dependency", "package": "sglang[all]"}],
         ),
-        (
-            r"No module named ['\"]?mlx_lm|mlx_lm.*command not found|MLX is not installed|MLX LM is not installed",
-            "MLX LM is not installed on this server.",
-            [{"label": "install mlx-lm in Cookbook Dependencies", "op": "dependency", "package": "mlx-lm"}],
-        ),
-        (
-            r"Unable to quantize model of type <class ['\"]mlx_lm\.models\.switch_layers\.QuantizedSwitchLinear['\"]>|QuantizedSwitchLinear",
-            "MLX-LM tried to quantize an already-quantized DeepSeek switch layer.",
-            [
-                {"label": "relaunch from the cached local Hugging Face snapshot path on this Mac", "op": "manual"},
-                {"label": "Odysseus now rewrites MLX repo-id launches to a cached snapshot when one exists", "op": "manual"},
-            ],
-        ),
         # System build deps come BEFORE the generic llama.cpp catch-all so
         # cmake / build-essential / git missing → a specific OS-package
         # remediation instead of "install llama-cpp-python[server]" (which

routes/document_helpers.py

@@ -29,7 +29,6 @@ class DocumentCreate(BaseModel):
 class DocumentUpdate(BaseModel):
     content: str
     summary: Optional[str] = None
-    force_version: bool = False
 
 class DocumentPatch(BaseModel):
     title: Optional[str] = None

routes/document_routes.py

@@ -54,18 +54,6 @@ def _library_language_for_document(doc: Document) -> str:
     return doc.language or "text"
 
 
-def _email_source_key(content: str) -> tuple[str, str]:
-    """Return the source email identity embedded in an email draft document."""
-    import re
-
-    text = content or ""
-    uid_m = re.search(r"(?im)^X-Source-UID:\s*(.+?)\s*$", text)
-    folder_m = re.search(r"(?im)^X-Source-Folder:\s*(.+?)\s*$", text)
-    uid = (uid_m.group(1).strip() if uid_m else "")
-    folder = (folder_m.group(1).strip() if folder_m else "INBOX")
-    return uid, folder
-
-
 from routes.document_helpers import (
     DocumentCreate, DocumentUpdate, DocumentPatch,
     _doc_to_dict, _version_to_dict,
@@ -111,62 +99,24 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                 # the existing lenient path.
                 session = _get_session_or_404(db, req.session_id, user)
 
+            doc_id = str(uuid.uuid4())
+            ver_id = str(uuid.uuid4())
+
             # If no language was supplied (e.g. cloning a doc whose language
             # was never set), detect it from the content rather than storing
             # NULL — which made the editor fall back to plain text. Defaults
             # to markdown for prose.
             language = req.language
             if not language:
-                from src.agent_tools.document_tools import _looks_like_email_document, _sniff_doc_language, _coerce_email_document_content
+                from src.agent_tools.document_tools import _looks_like_email_document, _sniff_doc_language
                 language = _sniff_doc_language(req.content)
             else:
-                from src.agent_tools.document_tools import _looks_like_email_document, _coerce_email_document_content
+                from src.agent_tools.document_tools import _looks_like_email_document
             if _looks_like_email_document(req.content, req.title):
                 language = "email"
 
             _assert_pdf_marker_upload_owned(request, req.content, user, upload_handler)
 
-            # Reply drafts are keyed to the source email. If a UI/tool path tries
-            # to create a second draft for the same email in the same chat,
-            # update the existing draft instead so quoted thread history stays
-            # attached to the visible document.
-            if language == "email" and req.session_id:
-                source_uid, source_folder = _email_source_key(req.content)
-                if source_uid:
-                    candidates = (
-                        db.query(Document)
-                        .filter(Document.session_id == req.session_id)
-                        .filter(Document.is_active == True)
-                        .filter(Document.language == "email")
-                        .order_by(Document.updated_at.desc())
-                        .limit(25)
-                        .all()
-                    )
-                    for existing in candidates:
-                        old_uid, old_folder = _email_source_key(existing.current_content or "")
-                        if old_uid != source_uid or old_folder != source_folder:
-                            continue
-                        merged = _coerce_email_document_content(existing.current_content or "", req.content)
-                        if existing.current_content != merged:
-                            new_ver = (existing.version_count or 1) + 1
-                            existing.current_content = merged
-                            existing.title = req.title or existing.title
-                            existing.version_count = new_ver
-                            db.add(DocumentVersion(
-                                id=str(uuid.uuid4()),
-                                document_id=existing.id,
-                                version_number=new_ver,
-                                content=merged,
-                                summary="Updated existing email draft",
-                                source="user",
-                            ))
-                            db.commit()
-                            db.refresh(existing)
-                        return _doc_to_dict(existing)
-
-            doc_id = str(uuid.uuid4())
-            ver_id = str(uuid.uuid4())
-
             doc = Document(
                 id=doc_id,
                 session_id=req.session_id,
@@ -620,23 +570,11 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                 raise HTTPException(404, "Document not found")
             _verify_doc_owner(db, doc, user)
 
-            incoming_content = req.content
-            from src.agent_tools.document_tools import _coerce_email_document_content, _looks_like_email_document
-            is_email_doc = (
-                (doc.language or "").lower() == "email"
-                or _looks_like_email_document(doc.current_content or "", doc.title or "")
-                or _looks_like_email_document(req.content or "", doc.title or "")
-            )
-            if is_email_doc:
-                incoming_content = _coerce_email_document_content(doc.current_content or "", req.content)
-                doc.language = "email"
-
-            # Skip if content is identical unless the caller explicitly wants
-            # a checkpoint version from the current editor state.
-            if doc.current_content == incoming_content and not req.force_version:
+            # Skip if content is identical
+            if doc.current_content == req.content:
                 return _doc_to_dict(doc)
 
-            _assert_pdf_marker_upload_owned(request, incoming_content, user, upload_handler)
+            _assert_pdf_marker_upload_owned(request, req.content, user, upload_handler)
 
             # Check if we can coalesce with the latest version
             latest_ver = db.query(DocumentVersion).filter(
@@ -645,14 +583,14 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
 
             now = datetime.now(timezone.utc)
             coalesced = False
-            if latest_ver and latest_ver.source == "user" and not req.force_version:
+            if latest_ver and latest_ver.source == "user":
                 ver_time = latest_ver.created_at
                 if ver_time.tzinfo is None:
                     ver_time = ver_time.replace(tzinfo=timezone.utc)
                 age = (now - ver_time).total_seconds()
                 if age < VERSION_COALESCE_SECONDS:
                     # Update the existing version in-place
-                    latest_ver.content = incoming_content
+                    latest_ver.content = req.content
                     latest_ver.created_at = now
                     if req.summary:
                         latest_ver.summary = req.summary
@@ -664,14 +602,14 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                     id=str(uuid.uuid4()),
                     document_id=doc_id,
                     version_number=new_ver,
-                    content=incoming_content,
+                    content=req.content,
                     summary=req.summary or "Manual edit",
                     source="user",
                 )
                 doc.version_count = new_ver
                 db.add(ver)
 
-            doc.current_content = incoming_content
+            doc.current_content = req.content
             db.commit()
             db.refresh(doc)
             return _doc_to_dict(doc)
@@ -861,26 +799,10 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
             from src.document_actions import _JUNK_TITLES
 
             to_delete = []
-            now = datetime.now(timezone.utc)
             for doc in docs:
-                created = doc.created_at
-                if created and created.tzinfo is None:
-                    created = created.replace(tzinfo=timezone.utc)
-
-                # Skip freshly created documents to avoid deleting them while the user is actively editing
-                if created and (now - created).total_seconds() < 900:  # 15 minutes
-                    continue
-
                 content = (doc.current_content or "").strip()
                 title_raw = (doc.title or "").strip()
                 title = title_raw.lower()
-                is_fresh_empty = (
-                    not content
-                    and created is not None
-                    and (now - created).total_seconds() < 1800
-                )
-                if is_fresh_empty:
-                    continue
 
                 # Strip markdown noise to get a "real" character count
                 stripped = _re.sub(r"^#{1,6}\s+", "", content, flags=_re.MULTILINE)
@@ -915,6 +837,10 @@ def setup_document_routes(session_manager, upload_handler=None) -> APIRouter:
                     to_delete.append(doc); deleted += 1; continue
                 if title in _JUNK_TITLES:
                     to_delete.append(doc); deleted += 1; continue
+                if real_len < 30:
+                    to_delete.append(doc); deleted += 1; continue
+                if "\n" not in content and real_len < 50:
+                    to_delete.append(doc); deleted += 1; continue
 
                 # Fix empty or placeholder titles on survivors
                 if not title_raw or title_raw == "Untitled":

routes/email_helpers.py

@@ -40,16 +40,6 @@ from src.secret_storage import decrypt as _decrypt
 logger = logging.getLogger(__name__)
 
 
-class EmailNotConfiguredError(RuntimeError):
-    """Raised when an IMAP operation is attempted on an account that has no
-    inbox configured (e.g. a send-only / SMTP-only account).
-
-    Subclasses RuntimeError so existing broad ``except Exception`` handlers
-    keep working; callers that want to treat "no inbox" as an empty result
-    rather than a failure can catch this type specifically.
-    """
-
-
 def _xoauth2_raw(user: str, access_token: str) -> str:
     """The SASL XOAUTH2 initial-response string (unencoded).
 
@@ -235,9 +225,8 @@ def _strip_think(text: str) -> str:
     """
     if not text:
         return ""
-    from src.text_helpers import strip_think as _central, _THINK_TAG_RE
-    # Single linear tag check; the old closed/open `.search()` calls could ReDoS.
-    had_think = bool(_THINK_TAG_RE.search(text))
+    from src.text_helpers import strip_think as _central, _THINK_CLOSED_RE, _THINK_OPEN_RE, _THINK_TAG_RE
+    had_think = bool(_THINK_CLOSED_RE.search(text) or _THINK_OPEN_RE.search(text) or _THINK_TAG_RE.search(text))
     return _central(text, prose=had_think, prompt_echo=True)
 
 
@@ -349,7 +338,7 @@ def _assert_owns_account(account_id: str, owner: str) -> None:
             row = db.query(_EA).filter(_EA.id == account_id).first()
             if row is None:
                 raise HTTPException(404, "Account not found")
-            if not _account_visible_to_owner(row, owner):
+            if row.owner and row.owner != owner:
                 # Treat as 404 (not 403) so we don't leak existence.
                 raise HTTPException(404, "Account not found")
         finally:
@@ -362,26 +351,6 @@ def _assert_owns_account(account_id: str, owner: str) -> None:
         logger.error(f"Account-owner check failed: {e}")
         raise HTTPException(503, "Account check failed")
 
-
-def _account_visible_to_owner(row, owner: str) -> bool:
-    """Whether an authenticated `owner` may act on this EmailAccount row.
-
-    Mirrors the SQL predicate in `_get_email_config`'s
-    `_owner_or_matching_legacy_account`: a caller sees an account they own, or a
-    legacy owner-less account (owner NULL/"") only when its own mailbox
-    (`imap_user` / `from_address`) is the caller's. `email_accounts` is the one
-    owner-scoped table deliberately left out of the legacy-owner migration
-    backfill, so ownerless rows persist on multi-user deploys — making this the
-    gate that keeps one tenant off another's imported mailbox and its decrypted
-    IMAP/SMTP credentials."""
-    row_owner = getattr(row, "owner", None) or ""
-    if row_owner:
-        return row_owner == owner
-    return owner in {
-        getattr(row, "imap_user", None) or "",
-        getattr(row, "from_address", None) or "",
-    }
-
 def _q(name: str) -> str:
     """Quote an IMAP mailbox name. Defensive: escapes `\\` and `"` and wraps
     in double quotes so user-supplied folder names with spaces or quotes can't
@@ -444,19 +413,12 @@ SCHEDULED_DB = Path(SCHEDULED_EMAILS_DB)
 OWNER_SCOPED_EMAIL_CACHE_TABLES = {
     "email_summaries",
     "email_ai_replies",
-    "email_translations",
     "email_calendar_extractions",
     "email_urgency_alerts",
     "sender_signatures",
 }
 
 
-def email_translation_body_hash(body: str) -> str:
-    import hashlib as _hashlib
-    normalized = (body or "").strip()
-    return _hashlib.sha256(normalized.encode("utf-8", errors="ignore")).hexdigest()
-
-
 def _email_cache_owner_clause(owner: str = "") -> tuple[str, tuple[str, ...]]:
     owner = (owner or "").strip()
     if owner:
@@ -464,34 +426,14 @@ def _email_cache_owner_clause(owner: str = "") -> tuple[str, tuple[str, ...]]:
     return "(owner = '' OR owner IS NULL)", ()
 
 
-def _ensure_owner_scoped_email_cache_table(
-    conn,
-    table: str,
-    create_sql: str,
-    columns: list[str],
-    pk_columns: list[str] | None = None,
-):
+def _ensure_owner_scoped_email_cache_table(conn, table: str, create_sql: str, columns: list[str]):
     """Rebuild legacy Message-ID-only cache tables with owner in the PK."""
-    desired_pk_cols = pk_columns or ["message_id", "owner"]
     conn.execute(create_sql)
     try:
         info = conn.execute(f"PRAGMA table_info({table})").fetchall()
         cols = [r[1] for r in info]
         pk_cols = [r[1] for r in sorted((r for r in info if r[5]), key=lambda r: r[5])]
-        for col in columns:
-            if col not in cols:
-                if col == "owner":
-                    conn.execute(f"ALTER TABLE {table} ADD COLUMN owner TEXT DEFAULT ''")
-                elif col in {"event_uids"}:
-                    conn.execute(f"ALTER TABLE {table} ADD COLUMN {col} TEXT DEFAULT '[]'")
-                elif col.startswith("has_") or col.endswith("_created") or col.endswith("_count"):
-                    conn.execute(f"ALTER TABLE {table} ADD COLUMN {col} INTEGER DEFAULT 0")
-                elif col == "created_at":
-                    conn.execute(f"ALTER TABLE {table} ADD COLUMN {col} TEXT DEFAULT ''")
-                else:
-                    conn.execute(f"ALTER TABLE {table} ADD COLUMN {col} TEXT")
-                cols.append(col)
-        if "owner" in cols and pk_cols == desired_pk_cols:
+        if "owner" in cols and pk_cols == ["message_id", "owner"]:
             return
 
         conn.execute(f"ALTER TABLE {table} RENAME TO {table}__old")
@@ -624,25 +566,6 @@ def _init_scheduled_db():
             PRIMARY KEY (message_id, owner)
         )
     """, ["message_id", "owner", "uid", "folder", "reply", "model_used", "created_at"])
-    _ensure_owner_scoped_email_cache_table(conn, "email_translations", """
-        CREATE TABLE IF NOT EXISTS email_translations (
-            body_hash TEXT,
-            owner TEXT DEFAULT '',
-            target_language TEXT DEFAULT 'English',
-            uid TEXT,
-            folder TEXT,
-            subject TEXT,
-            sender TEXT,
-            translation TEXT,
-            same_language INTEGER DEFAULT 0,
-            model_used TEXT,
-            created_at TEXT NOT NULL,
-            PRIMARY KEY (body_hash, owner, target_language)
-        )
-    """, [
-        "body_hash", "owner", "target_language", "uid", "folder", "subject", "sender",
-        "translation", "same_language", "model_used", "created_at",
-    ], ["body_hash", "owner", "target_language"])
     # Email tags / spam classification cache. SECURITY: keyed by
     # (message_id, owner) because Message-IDs are GLOBAL (a newsletter goes
     # to many users with the same Message-ID). Without owner-scoping, a
@@ -652,7 +575,6 @@ def _init_scheduled_db():
         CREATE TABLE IF NOT EXISTS email_tags (
             message_id TEXT,
             owner TEXT DEFAULT '',
-            account_id TEXT DEFAULT '',
             uid TEXT,
             folder TEXT,
             subject TEXT,
@@ -663,7 +585,7 @@ def _init_scheduled_db():
             moved_to TEXT,
             model_used TEXT,
             created_at TEXT NOT NULL,
-            PRIMARY KEY (message_id, owner, account_id)
+            PRIMARY KEY (message_id, owner)
         )
     """)
     # Backfill migration: older installs created the table with
@@ -671,35 +593,28 @@ def _init_scheduled_db():
     # promote it into the PK by rebuild-copy-swap (SQLite can't ALTER PK).
     try:
         _cols = [r[1] for r in conn.execute("PRAGMA table_info(email_tags)")]
-        _pk_cols = [r[1] for r in sorted(conn.execute("PRAGMA table_info(email_tags)").fetchall(), key=lambda row: row[5] or 99) if r[5]]
         if "owner" not in _cols:
+            # Add the column first so reads/writes don't break mid-migration.
             conn.execute("ALTER TABLE email_tags ADD COLUMN owner TEXT DEFAULT ''")
-            _cols.append("owner")
-        if "account_id" not in _cols:
-            conn.execute("ALTER TABLE email_tags ADD COLUMN account_id TEXT DEFAULT ''")
-            _cols.append("account_id")
-        if _pk_cols != ["message_id", "owner", "account_id"]:
-            # Rebuild with account-aware composite PK. Existing rows get
-            # account_id='' and are still readable as legacy fallback rows;
-            # fresh task runs write exact account ids and no longer block each
-            # other when two accounts share a Message-ID.
+            # Rebuild with composite PK. Existing rows get owner='' (legacy
+            # single-user); the urgency scanner will overwrite as it
+            # re-classifies. No data loss.
             conn.execute("""
                 CREATE TABLE IF NOT EXISTS email_tags__new (
                     message_id TEXT,
                     owner TEXT DEFAULT '',
-                    account_id TEXT DEFAULT '',
                     uid TEXT, folder TEXT, subject TEXT, sender TEXT,
                     tags TEXT, spam_verdict INTEGER DEFAULT 0,
                     spam_reason TEXT, moved_to TEXT, model_used TEXT,
                     created_at TEXT NOT NULL,
-                    PRIMARY KEY (message_id, owner, account_id)
+                    PRIMARY KEY (message_id, owner)
                 )
             """)
             conn.execute("""
                 INSERT OR IGNORE INTO email_tags__new
-                  (message_id, owner, account_id, uid, folder, subject, sender, tags,
+                  (message_id, owner, uid, folder, subject, sender, tags,
                    spam_verdict, spam_reason, moved_to, model_used, created_at)
-                SELECT message_id, COALESCE(owner, ''), COALESCE(account_id, ''), uid, folder, subject,
+                SELECT message_id, COALESCE(owner, ''), uid, folder, subject,
                        sender, tags, spam_verdict, spam_reason, moved_to,
                        model_used, created_at
                 FROM email_tags
@@ -715,12 +630,11 @@ def _init_scheduled_db():
             message_id TEXT,
             owner TEXT DEFAULT '',
             uid TEXT,
-            event_uids TEXT DEFAULT '[]',
             events_created INTEGER DEFAULT 0,
             created_at TEXT NOT NULL,
             PRIMARY KEY (message_id, owner)
         )
-    """, ["message_id", "owner", "uid", "event_uids", "events_created", "created_at"])
+    """, ["message_id", "owner", "uid", "events_created", "created_at"])
     _ensure_owner_scoped_email_cache_table(conn, "email_urgency_alerts", """
         CREATE TABLE IF NOT EXISTS email_urgency_alerts (
             message_id TEXT,
@@ -746,64 +660,6 @@ def _init_scheduled_db():
             PRIMARY KEY (owner, account_key, folder, message_key)
         )
     """)
-    conn.execute("""
-        CREATE TABLE IF NOT EXISTS email_message_index (
-            owner TEXT NOT NULL DEFAULT '',
-            account_key TEXT NOT NULL DEFAULT '',
-            folder TEXT NOT NULL,
-            uid TEXT NOT NULL,
-            message_id TEXT,
-            subject TEXT,
-            from_name TEXT,
-            from_address TEXT,
-            to_text TEXT,
-            cc_text TEXT,
-            date_iso TEXT,
-            date_display TEXT,
-            date_epoch REAL DEFAULT 0,
-            size INTEGER DEFAULT 0,
-            flags TEXT DEFAULT '',
-            has_attachments INTEGER DEFAULT 0,
-            updated_at TEXT NOT NULL,
-            PRIMARY KEY (owner, account_key, folder, uid)
-        )
-    """)
-    conn.execute("""
-        CREATE INDEX IF NOT EXISTS ix_email_message_index_folder_date
-        ON email_message_index(owner, account_key, folder, date_epoch DESC)
-    """)
-    conn.execute("""
-        CREATE INDEX IF NOT EXISTS ix_email_message_index_message_id
-        ON email_message_index(owner, account_key, message_id)
-    """)
-    conn.execute("""
-        CREATE TABLE IF NOT EXISTS email_body_preview_cache (
-            owner TEXT NOT NULL DEFAULT '',
-            account_key TEXT NOT NULL DEFAULT '',
-            folder TEXT NOT NULL,
-            uid TEXT NOT NULL,
-            message_id TEXT,
-            payload_json TEXT NOT NULL,
-            updated_at TEXT NOT NULL,
-            PRIMARY KEY (owner, account_key, folder, uid)
-        )
-    """)
-    conn.execute("""
-        CREATE INDEX IF NOT EXISTS ix_email_body_preview_message_id
-        ON email_body_preview_cache(owner, account_key, message_id)
-    """)
-    conn.execute("""
-        CREATE TABLE IF NOT EXISTS email_attachment_metadata_cache (
-            owner TEXT NOT NULL DEFAULT '',
-            account_key TEXT NOT NULL DEFAULT '',
-            folder TEXT NOT NULL,
-            uid TEXT NOT NULL,
-            message_id TEXT,
-            attachments_json TEXT NOT NULL,
-            updated_at TEXT NOT NULL,
-            PRIMARY KEY (owner, account_key, folder, uid)
-        )
-    """)
     # Boundary cache — LLM-detected sig/quote start positions in the body.
     # Stored as char offsets (-1 = no boundary found). Once cached, the
     # client uses these to fold without ever re-calling the LLM.
@@ -923,13 +779,12 @@ def _get_email_config(account_id: str | None = None, owner: str = "") -> dict:
         try:
             if account_id:
                 row = db.query(_EA).filter(_EA.id == account_id, _EA.enabled == True).first()  # noqa: E712
-                # If the resolved row isn't visible to this owner, treat as
+                # If the resolved row belongs to a different owner, treat as
                 # not-found rather than silently serving it. This is a defense
                 # in depth — `require_owner` already calls `_assert_owns_account`
                 # for query-param account_ids, but other callers (cookbook
-                # rules, scheduled poller) may not. Ownerless legacy rows are
-                # only visible on a mailbox match, same as the fallback below.
-                if row is not None and owner and not _account_visible_to_owner(row, owner):
+                # rules, scheduled poller) may not.
+                if row is not None and owner and row.owner and row.owner != owner:
                     row = None
             # Fallback path — restrict to this owner's accounts so we don't
             # leak another user's default mailbox to an unconfigured user.
@@ -1073,14 +928,6 @@ def _imap_connect(account_id: str | None = None, owner: str = "",
     # `timeout` is overridable so short-lived callers (e.g. the service-health
     # probe) can impose a tighter budget than the default IMAP timeout.
     cfg = _get_email_config(account_id, owner=owner)
-    # Send-only (SMTP-only) account: no IMAP host means there is no inbox to
-    # read. Bail out with a clear, typed error instead of handing an empty
-    # host to imaplib — IMAP4("", 993) silently dials localhost:993 and fails
-    # with a confusing "[Errno 111] Connection refused" on every inbox poll.
-    if not cfg.get("imap_host"):
-        raise EmailNotConfiguredError(
-            f"IMAP is not configured for account {cfg.get('account_name') or 'default'!r}"
-        )
     # Connection mode:
     #   STARTTLS on → plain + upgrade
     #   STARTTLS off + port 993 → implicit SSL (IMAPS)
@@ -1294,15 +1141,10 @@ def _imap_move(uid, dest, src="INBOX", account_id: str | None = None, owner: str
     try:
         c = _imap_connect(account_id, owner=owner)
         c.select(_q(src))
-        # Callers pass a real IMAP UID (from conn.uid("SEARCH", ...)). copy()
-        # and store() operate on message SEQUENCE NUMBERS, so addressing them
-        # with a UID moved/deleted the wrong message (or silently no-oped when
-        # the UID exceeded the message count). Use the UID commands, matching
-        # the move/delete path in email_routes.py.
-        status, _ = c.uid("COPY", uid, _q(dest))
+        status, _ = c.copy(uid, _q(dest))
         if status != "OK":
             return False
-        c.uid("STORE", uid, "+FLAGS", "\\Deleted")
+        c.store(uid, "+FLAGS", "\\Deleted")
         c.expunge()
         return True
     except Exception as e:
@@ -1415,14 +1257,12 @@ def _list_attachments_from_msg(msg):
             except Exception:
                 payload = b""
         size = len(payload) if payload is not None else 0
-        content_id = (part.get("Content-ID") or "").strip().strip("<>")
         attachments.append({
             "index": idx,
             "filename": filename,
             "content_type": ct,
             "size": size,
             "is_inline": "inline" in cd.lower(),
-            "content_id": content_id,
         })
         idx += 1
     return attachments
@@ -1861,10 +1701,6 @@ class SendEmailRequest(BaseModel):
     attachments: Optional[List[str]] = None
     # Which account to send from. None = default account.
     account_id: Optional[str] = None
-    # Source message for replies. When present, /send marks this exact message
-    # answered after successful delivery so it leaves undone/reply-soon views.
-    source_uid: Optional[str] = None
-    source_folder: Optional[str] = None
     # Internal marker for Odysseus-generated mail (e.g. reminder, scheduled).
     odysseus_kind: Optional[str] = None
     # If true, /send waits for SMTP + Sent append and returns the sent UID.

routes/email_pollers.py

@@ -29,7 +29,7 @@ from datetime import datetime
 from email.mime.text import MIMEText
 from email.mime.multipart import MIMEMultipart
 
-from src.task_endpoint import resolve_task_candidates, task_llm_call_async
+from src.llm_core import llm_call_async
 
 from routes.email_helpers import (
     _strip_think, _extract_reply, _apply_email_style_mechanics, _load_settings, _save_settings, _get_email_config,
@@ -56,35 +56,6 @@ _CAL_ACTION_ARRAY_RE = re.compile(
 )
 
 
-def _extract_json_array_from_text(text: str):
-    """Return the last valid JSON array embedded in model output, if any."""
-    if not text:
-        return None
-    cleaned = re.sub(r"^```(?:json)?\s*|\s*```$", "", text.strip(), flags=re.MULTILINE).strip()
-    decoder = json.JSONDecoder()
-    try:
-        parsed = decoder.decode(cleaned)
-        if isinstance(parsed, list):
-            return parsed
-    except Exception:
-        pass
-
-    # Models often explain themselves and finish with `[]` or `[{"action":...}]`.
-    # Scan every array opener and keep the last complete JSON array, rather than
-    # using a greedy regex that can swallow prose containing square brackets.
-    last = None
-    for idx, ch in enumerate(cleaned):
-        if ch != "[":
-            continue
-        try:
-            parsed, _end = decoder.raw_decode(cleaned[idx:])
-        except Exception:
-            continue
-        if isinstance(parsed, list):
-            last = parsed
-    return last
-
-
 def _owner_for_email_account(account_id: str | None) -> str:
     if not account_id:
         return ""
@@ -117,8 +88,6 @@ async def _run_auto_summarize_once(do_summary: bool = True, do_reply: bool = Tru
                                    do_tag: bool = False, do_spam: bool = False,
                                    do_calendar: bool = False,
                                    days_back: int = 1,
-                                   account_id: str | None = None,
-                                   max_process: int | None = None,
                                    progress_cb=None) -> str:
     """One iteration of the email scan. Temporarily flips settings flags
     so the existing background-loop logic runs exactly once for the requested ops."""
@@ -133,12 +102,7 @@ async def _run_auto_summarize_once(do_summary: bool = True, do_reply: bool = Tru
     settings["email_auto_calendar"] = bool(do_calendar)
     _save_settings(settings)
     try:
-        return await _auto_summarize_pass(
-            days_back=days_back,
-            account_id=account_id,
-            max_process=max_process,
-            progress_cb=progress_cb,
-        )
+        return await _auto_summarize_pass(days_back=days_back, progress_cb=progress_cb)
     finally:
         s2 = _load_settings()
         for k, v in prev.items():
@@ -176,7 +140,7 @@ def _latest_inbox_fallback_uids(conn, reconnect):
         return [], reconnect()
 
 
-async def _auto_summarize_pass(days_back: int = 1, account_id: str | None = None, max_process: int | None = None, progress_cb=None) -> str:
+async def _auto_summarize_pass(days_back: int = 1, account_id: str | None = None, progress_cb=None) -> str:
     """Single pass of the auto-summarize/reply scan.
 
     When account_id is None, iterates over every enabled account in
@@ -203,41 +167,28 @@ async def _auto_summarize_pass(days_back: int = 1, account_id: str | None = None
             names = {}
         if len(ids) <= 1:
             # Single-account (or zero rows — fallback to legacy settings.json lookup)
-            return await _auto_summarize_pass_single(
-                days_back=days_back,
-                account_id=(ids[0] if ids else None),
-                max_process=max_process,
-                progress_cb=progress_cb,
-            )
+            return await _auto_summarize_pass_single(days_back=days_back, account_id=(ids[0] if ids else None), progress_cb=progress_cb)
         outs = []
         for idx, aid in enumerate(ids, start=1):
             try:
                 await _emit_progress(progress_cb, f"{names.get(aid, aid[:8])}: starting ({idx}/{len(ids)})")
-                result = await _auto_summarize_pass_single(
-                    days_back=days_back,
-                    account_id=aid,
-                    max_process=max_process,
-                    progress_cb=progress_cb,
-                )
+                result = await _auto_summarize_pass_single(days_back=days_back, account_id=aid, progress_cb=progress_cb)
                 outs.append(f"[{names.get(aid, aid[:8])}] {result}")
             except Exception as e:
                 logger.warning(f"auto-summarize pass failed for account {aid}: {e}")
                 outs.append(f"[{names.get(aid, aid[:8])}] error: {e}")
         return "\n".join(outs)
-    return await _auto_summarize_pass_single(
-        days_back=days_back,
-        account_id=account_id,
-        max_process=max_process,
-        progress_cb=progress_cb,
-    )
+    return await _auto_summarize_pass_single(days_back=days_back, account_id=account_id, progress_cb=progress_cb)
 
 
-async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None = None, max_process: int | None = None, progress_cb=None) -> str:
+async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None = None, progress_cb=None) -> str:
     """Single pass of the auto-summarize/reply scan for ONE account.
     Reads current settings flags."""
     import asyncio
     import sqlite3 as _sql3
-    from src.llm_core import _uses_max_completion_tokens
+    import requests as _req
+    from src.endpoint_resolver import resolve_endpoint
+    from src.llm_core import _uses_max_completion_tokens, _restricts_temperature
 
     settings = _load_settings()
     auto_sum = settings.get("email_auto_summarize", False)
@@ -314,15 +265,9 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
         ).fetchall()}
         if auto_tag or auto_spam:
             if account_owner:
-                _tag_existing = {r[0] for r in _c.execute(
-                    "SELECT message_id FROM email_tags WHERE owner=? AND (account_id=? OR account_id='' OR account_id IS NULL)",
-                    (account_owner, account_id or ""),
-                ).fetchall()}
+                _tag_existing = {r[0] for r in _c.execute("SELECT message_id FROM email_tags WHERE owner=?", (account_owner,)).fetchall()}
             else:
-                _tag_existing = {r[0] for r in _c.execute(
-                    "SELECT message_id FROM email_tags WHERE (owner='' OR owner IS NULL) AND (account_id=? OR account_id='' OR account_id IS NULL)",
-                    (account_id or "",),
-                ).fetchall()}
+                _tag_existing = {r[0] for r in _c.execute("SELECT message_id FROM email_tags WHERE owner='' OR owner IS NULL").fetchall()}
         else:
             _tag_existing = set()
         _cal_existing = {r[0] for r in _c.execute(
@@ -351,10 +296,11 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
         if auto_spam and not spam_folder:
             logger.warning("Auto-spam enabled but no Junk/Spam folder detected — will classify but not move")
 
-        task_candidates = resolve_task_candidates(owner=account_owner)
-        if not task_candidates:
+        url, model, headers = resolve_endpoint("utility", owner=account_owner)
+        if not url:
+            url, model, headers = resolve_endpoint("default", owner=account_owner)
+        if not url or not model:
             return "No model configured"
-        url, model, headers = task_candidates[0]
 
         writing_style = settings.get("email_writing_style", "")
         processed = 0
@@ -368,14 +314,7 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
         _reply_failed = 0
         _detail_lines = []
         _current_folder = "INBOX"
-        # Calendar extraction is sequential and each row can involve a model
-        # call plus a calendar write. Keep the scheduled calendar-only pass
-        # below the 5-minute action budget instead of timing out mid-run.
-        _default_max_process = 3 if (auto_cal and not auto_sum and not auto_reply and not auto_tag and not auto_spam) else 5
-        try:
-            _max_process = max(1, int(max_process)) if max_process is not None else _default_max_process
-        except Exception:
-            _max_process = _default_max_process
+        _max_process = 5
         for _entry in uid_list:
             if processed >= _max_process:
                 break
@@ -467,30 +406,48 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                     req_headers.update(headers)
 
                 if need_sum:
+                    tok_key = "max_completion_tokens" if _uses_max_completion_tokens(model) else "max_tokens"
+                    payload = {
+                        "model": model,
+                        "messages": [
+                            {"role": "system", "content": "You are an email summarizer. Format: 1-3 short bullet points (use '- '). Cover: main point, action items, deadlines. If the email has attachments (marked '--- ATTACHMENTS ---'), USE THEIR CONTENTS — pull out invoice totals, deadlines, key clauses, any concrete numbers/dates in PDFs/docs, and reflect them in the bullets. Be terse.\n\nOUTPUT FORMAT: Put ONLY the bullet points between these exact markers, each on its own line:\n<<<SUMMARY>>>\n- ...\n<<<END>>>\nAny reasoning or planning must come BEFORE <<<SUMMARY>>> (ideally inside <think>...</think>). Only the text between the markers is kept."},
+                            {"role": "user", "content": f"From: {sender}\nSubject: {subject}\n\n{body_for_llm[:12000]}\n\n---\n\nSummarize the email. Output the bullets between <<<SUMMARY>>> and <<<END>>>."},
+                        ],
+                        tok_key: 16384,
+                        "temperature": 0.3,
+                        "stream": False,
+                    }
+                    # Reasoning models (o1/o3/o4/gpt-5) reject an explicit temperature.
+                    if _restricts_temperature(model):
+                        payload.pop("temperature", None)
                     try:
-                        summary = await task_llm_call_async(
-                            messages=[
-                                {"role": "system", "content": "You are an email summarizer. Format: 1-3 short bullet points (use '- '). Cover: main point, action items, deadlines. If the email has attachments (marked '--- ATTACHMENTS ---'), USE THEIR CONTENTS — pull out invoice totals, deadlines, key clauses, any concrete numbers/dates in PDFs/docs, and reflect them in the bullets. Be terse.\n\nOUTPUT FORMAT: Put ONLY the bullet points between these exact markers, each on its own line:\n<<<SUMMARY>>>\n- ...\n<<<END>>>\nAny reasoning or planning must come BEFORE <<<SUMMARY>>> (ideally inside <think>...</think>). Only the text between the markers is kept."},
-                                {"role": "user", "content": f"From: {sender}\nSubject: {subject}\n\n{body_for_llm[:12000]}\n\n---\n\nSummarize the email. Output the bullets between <<<SUMMARY>>> and <<<END>>>."},
-                            ],
-                            fallback_url=url, fallback_model=model, fallback_headers=headers,
-                            owner=account_owner or None,
-                            temperature=0.3, max_tokens=16384, timeout=240,
+                        # Use to_thread so this sync HTTP call doesn't freeze
+                        # the entire event loop while the LLM thinks (240s).
+                        resp = await asyncio.to_thread(
+                            _req.post, url, json=payload, headers=req_headers, timeout=240
                         )
-                        summary = _extract_reply((summary or "").strip())
-                        if summary:
-                            _c = _sql3.connect(SCHEDULED_DB)
-                            _c.execute("""
-                                INSERT OR REPLACE INTO email_summaries
-                                (message_id, owner, uid, folder, subject, sender, summary, model_used, created_at)
-                                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
-                            """, (message_id, account_owner or "", uid.decode() if isinstance(uid, bytes) else str(uid), _folder, subject, sender, summary, model, datetime.utcnow().isoformat()))
-                            _c.commit()
-                            _c.close()
-                            _sum_existing.add(message_id)
-                            _summaries_created += 1
-                            _uid_text = uid.decode() if isinstance(uid, bytes) else str(uid)
-                            _detail_lines.append(f"summary · {_folder}#{_uid_text} · {subject or '(no subject)'} — {sender or '(unknown sender)'}")
+                        if resp.ok:
+                            rdata = resp.json()
+                            m = (rdata.get("choices") or [{}])[0].get("message", {})
+                            summary = (m.get("content") or "").strip()
+                            summary = _extract_reply(summary)
+                            if not summary:
+                                rc = (m.get("reasoning_content") or "").strip()
+                                bullets = [ln.strip() for ln in rc.split("\n") if re.match(r"^[-•*]\s+|^\d+[.)]\s+", ln.strip())]
+                                summary = "\n".join(bullets) if bullets else ""
+                            if summary:
+                                _c = _sql3.connect(SCHEDULED_DB)
+                                _c.execute("""
+                                    INSERT OR REPLACE INTO email_summaries
+                                    (message_id, owner, uid, folder, subject, sender, summary, model_used, created_at)
+                                    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+                                """, (message_id, account_owner or "", uid.decode() if isinstance(uid, bytes) else str(uid), _folder, subject, sender, summary, model, datetime.utcnow().isoformat()))
+                                _c.commit()
+                                _c.close()
+                                _sum_existing.add(message_id)
+                                _summaries_created += 1
+                                _uid_text = uid.decode() if isinstance(uid, bytes) else str(uid)
+                                _detail_lines.append(f"summary · {_folder}#{_uid_text} · {subject or '(no subject)'} — {sender or '(unknown sender)'}")
                     except Exception as e:
                         _uid_text = uid.decode() if isinstance(uid, bytes) else str(uid)
                         _detail_lines.append(f"summary failed · {_folder}#{_uid_text} · {subject or '(no subject)'} — {sender or '(unknown sender)'}")
@@ -511,14 +468,14 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                     if context_snippets:
                         sys_prompt += "\n\nRELEVANT CONTEXT FROM PAST EMAILS AND CONTACTS:\n" + "\n\n---\n\n".join(context_snippets[:5])
                     try:
-                        reply = await task_llm_call_async(
+                        reply = await llm_call_async(
+                            url=url, model=model,
                             messages=[
                                 {"role": "system", "content": sys_prompt},
                                 {"role": "user", "content": f"Original email:\nFrom: {sender}\nSubject: {subject}\n\n{body_for_llm[:12000]}\n\nDraft a reply. Return only the reply body text."},
                             ],
-                            fallback_url=url, fallback_model=model, fallback_headers=headers,
-                            owner=account_owner or None,
-                            temperature=0.7, max_tokens=1024, timeout=90,
+                            temperature=0.7, max_tokens=1024,
+                            headers=req_headers, timeout=90,
                         )
                         reply = _apply_email_style_mechanics(_extract_reply(reply or ""))
                         if reply:
@@ -545,8 +502,6 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                 # ── Calendar event extraction (independent of reply drafting) ──
                 if need_cal:
                     _cal_run_count = 0
-                    _cal_event_uids = []
-                    _cal_parse_ok = False
                     try:
                         # Pull a snapshot of upcoming events so the LLM can decide
                         # create vs update vs cancel based on what already exists.
@@ -555,7 +510,8 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                         _existing_summary = get_upcoming_events(_acct_owner, horizon_days=60, limit=40)
                         existing_json = json.dumps(_existing_summary)
                         is_sent = _folder.lower().startswith("sent") or "sent" in _folder.lower()
-                        cal_extract = await task_llm_call_async(
+                        cal_extract = await llm_call_async(
+                            url=url, model=model,
                             messages=[
                                 {"role": "system", "content": (
                                     "You are a calendar assistant. The user receives emails AND sends replies "
@@ -606,9 +562,8 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                                     f"{body[:4000]}"
                                 )},
                             ],
-                            fallback_url=url, fallback_model=model, fallback_headers=headers,
-                            owner=account_owner or None,
-                            temperature=0.1, max_tokens=16384, timeout=75,
+                            temperature=0.1, max_tokens=16384,
+                            headers=req_headers, timeout=180,
                         )
                         _raw_original = cal_extract or ""
                         cal_extract = _strip_think(_raw_original)
@@ -618,10 +573,10 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                             if matches:
                                 cal_extract = matches[-1].group()
                         logger.info(f"[cal-extract] uid={uid.decode() if isinstance(uid, bytes) else uid} folder={_folder} subj={subject[:50]!r} raw_len={len(cal_extract)} orig_len={len(_raw_original)} raw={cal_extract[:800]!r}")
-                        ops = _extract_json_array_from_text(cal_extract)
-                        if ops is not None:
+                        jm = re.search(r'\[.*\]', cal_extract, re.DOTALL)
+                        if jm:
                             try:
-                                _cal_parse_ok = True
+                                ops = json.loads(jm.group())
                                 logger.info(f"[cal-extract] parsed {len(ops)} op(s)")
                                 if isinstance(ops, list) and ops:
                                     from src.tool_implementations import do_manage_calendar
@@ -651,8 +606,6 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                                             r = await do_manage_calendar(json.dumps(args), owner=_acct_owner)
                                             if r.get("exit_code", 0) == 0:
                                                 logger.info(f"[cal-extract] Updated event uid={cuid} → {op.get('title')} {op['date']}")
-                                                if cuid and cuid not in _cal_event_uids:
-                                                    _cal_event_uids.append(cuid)
                                                 _cal_run_count += 1
                                             else:
                                                 logger.warning(f"[cal-extract] update failed: {r.get('error')}")
@@ -733,41 +686,29 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                                             r = await do_manage_calendar(cal_args, owner=_acct_owner)
                                             if r.get("exit_code", 0) == 0:
                                                 logger.info(f"[cal-extract] Created event: {op['title']} on {op['date']}")
-                                                _created_uid = (r.get("uid") or "").strip()
-                                                if _created_uid and _created_uid not in _cal_event_uids:
-                                                    _cal_event_uids.append(_created_uid)
                                                 _events_created += 1
                                                 _cal_run_count += 1
                                             else:
                                                 logger.warning(f"[cal-extract] create failed: {r.get('error')} args={cal_args[:200]}")
                             except Exception as je:
                                 logger.warning(f"[cal-extract] JSON parse failed: {je} on raw={cal_extract[:200]!r}")
-                        else:
-                            logger.warning(f"[cal-extract] no JSON array found on raw={cal_extract[:200]!r}")
                     except Exception as e:
                         logger.warning(f"[cal-extract] Meeting extraction LLM call failed for uid={uid}: {e}")
                     else:
-                        # Record successfully parsed results so we don't re-LLM
-                        # no-op emails. Transient LLM failures are retried on
-                        # the next poll run.
+                        # Record we processed this email so we don't re-LLM next run.
+                        # Only mark as processed on success ? transient LLM failures
+                        # are retried on the next poll run (matches summary/reply pattern).
                         try:
-                            if _cal_parse_ok:
-                                _cc = _sql3.connect(SCHEDULED_DB)
-                                _cc.execute(
-                                    "INSERT OR REPLACE INTO email_calendar_extractions "
-                                    "(message_id, owner, uid, event_uids, events_created, created_at) VALUES (?, ?, ?, ?, ?, ?)",
-                                    (
-                                        message_id,
-                                        account_owner or "",
-                                        uid.decode() if isinstance(uid, bytes) else str(uid),
-                                        json.dumps(_cal_event_uids),
-                                        _cal_run_count,
-                                        datetime.utcnow().isoformat(),
-                                    ),
-                                )
-                                _cc.commit()
-                                _cc.close()
-                                _cal_existing.add(message_id)
+                            _cc = _sql3.connect(SCHEDULED_DB)
+                            _cc.execute(
+                                "INSERT OR REPLACE INTO email_calendar_extractions "
+                                "(message_id, owner, uid, events_created, created_at) VALUES (?, ?, ?, ?, ?)",
+                                (message_id, account_owner or "", uid.decode() if isinstance(uid, bytes) else str(uid),
+                                 _cal_run_count, datetime.utcnow().isoformat())
+                            )
+                            _cc.commit()
+                            _cc.close()
+                            _cal_existing.add(message_id)
                         except Exception as ce:
                             logger.debug(f"Could not cache calendar extraction: {ce}")
 
@@ -801,11 +742,9 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                             "temperature": 0,
                             tok_key: 200,
                         }
-                        urg_raw = await task_llm_call_async(
-                            messages=payload["messages"],
-                            fallback_url=url, fallback_model=model, fallback_headers=headers,
-                            owner=account_owner or None,
-                            temperature=0, max_tokens=200, timeout=60,
+                        urg_raw = await llm_call_async(
+                            url=url, model=model, messages=payload["messages"],
+                            temperature=0, max_tokens=200, headers=req_headers, timeout=60,
                         )
                         urg_raw = _strip_think(urg_raw or "")
                         urg_raw = re.sub(r"^```(?:json)?\s*|\s*```$", "", urg_raw, flags=re.MULTILINE).strip()
@@ -906,13 +845,8 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                         class_sys = (
                             "Classify the email. Return ONLY a JSON object, no prose, no markdown fences. "
                             "Schema: {\"tags\": [\"tag1\"], \"spam\": false, \"reason\": \"short\"}. "
-                            "Pick 1-3 tags from: work, personal, urgent, action-needed, finance, bills, "
-                            "receipt, legal, travel, newsletter, promo, notification, security, social, "
-                            "shopping, calendar, support.\n\n"
-                            "Use work for professional/company/client/operations messages. "
-                            "Use personal for friends/family/private-life messages. "
-                            "Use urgent for real time-sensitive consequences. "
-                            "Use action-needed when the user likely needs to reply, pay, sign, book, or decide.\n\n"
+                            "Pick 1-2 tags from: work, personal, finance, bills, receipt, travel, "
+                            "newsletter, promo, notification, security, social, shopping, calendar.\n\n"
                             "Set spam=true for ANY of:\n"
                             "- Phishing, scams, chain mail, deceptive offers\n"
                             "- Marketing/promotional blasts (\"special offer\", \"limited time\", discount codes)\n"
@@ -929,55 +863,70 @@ async def _auto_summarize_pass_single(days_back: int = 1, account_id: str | None
                             "If it's a mass-mailed generic update with no personal CTA, mark spam=true even if from a legitimate service. "
                             "Reason should be 5-10 words."
                         )
-                        raw_out = await task_llm_call_async(
-                            messages=[
+                        tok_key = "max_completion_tokens" if _uses_max_completion_tokens(model) else "max_tokens"
+                        payload = {
+                            "model": model,
+                            "messages": [
                                 {"role": "system", "content": class_sys},
                                 {"role": "user", "content": f"From: {sender}\nSubject: {subject}\n\n{body[:4000]}"},
                             ],
-                            fallback_url=url, fallback_model=model, fallback_headers=headers,
-                            owner=account_owner or None,
-                            temperature=0.1, max_tokens=512, timeout=120,
+                            tok_key: 512,
+                            "temperature": 0.1,
+                            "stream": False,
+                        }
+                        # Reasoning models (o1/o3/o4/gpt-5) reject an explicit temperature.
+                        if _restricts_temperature(model):
+                            payload.pop("temperature", None)
+                        # to_thread keeps the event loop responsive during the LLM call
+                        resp = await asyncio.to_thread(
+                            _req.post, url, json=payload, headers=req_headers, timeout=120
                         )
-                        raw_out = _strip_think((raw_out or "").strip())
-                        raw_out = re.sub(r"^```(?:json)?\s*|\s*```$", "", raw_out, flags=re.MULTILINE).strip()
-                        jm = re.search(r'\{.*\}', raw_out, re.DOTALL)
-                        parsed = None
-                        if jm:
-                            try:
-                                parsed = json.loads(jm.group(0))
-                            except Exception:
-                                parsed = None
-                        if parsed is not None:
-                            _ALLOWED_TAGS = {"work","personal","urgent","action-needed","finance","bills",
-                                             "receipt","legal","travel","newsletter","marketing","notification",
-                                             "security","social","shopping","calendar","support"}
-                            raw_tags = parsed.get("tags") or []
-                            if isinstance(raw_tags, str):
-                                raw_tags = [raw_tags]
-                            tags = [t.strip().lower().replace("_", "-") for t in raw_tags if isinstance(t, str)]
-                            tags = ["marketing" if t == "promo" else t for t in tags]
-                            tags = [t for t in tags if t in _ALLOWED_TAGS][:3]
-                            is_spam = bool(parsed.get("spam"))
-                            spam_reason = str(parsed.get("reason") or "")[:200]
+                        if not resp.ok:
+                            logger.warning(f"Auto-classify {uid.decode() if isinstance(uid, bytes) else str(uid)} HTTP {resp.status_code}: {resp.text[:200]}")
+                        else:
+                            rdata = resp.json()
+                            m = (rdata.get("choices") or [{}])[0].get("message", {})
+                            raw_out = (m.get("content") or "").strip()
+                            raw_out = _strip_think(raw_out)
+                            raw_out = re.sub(r"^```(?:json)?\s*|\s*```$", "", raw_out, flags=re.MULTILINE).strip()
+                            jm = re.search(r'\{.*\}', raw_out, re.DOTALL)
+                            parsed = None
+                            if jm:
+                                try:
+                                    parsed = json.loads(jm.group(0))
+                                except Exception:
+                                    parsed = None
+                            if parsed is not None:
+                                _ALLOWED_TAGS = {"work","personal","finance","bills","receipt","travel",
+                                                 "newsletter","marketing","notification","security","social",
+                                                 "shopping","calendar"}
+                                raw_tags = parsed.get("tags") or []
+                                if isinstance(raw_tags, str):
+                                    raw_tags = [raw_tags]
+                                tags = [t.strip().lower().replace("_", "-") for t in raw_tags if isinstance(t, str)]
+                                tags = ["marketing" if t == "promo" else t for t in tags]
+                                tags = [t for t in tags if t in _ALLOWED_TAGS][:2]
+                                is_spam = bool(parsed.get("spam"))
+                                spam_reason = str(parsed.get("reason") or "")[:200]
 
-                            moved_to = ""
-                            if is_spam and auto_spam and spam_folder:
-                                if _imap_move(uid, spam_folder, account_id=account_id, owner=account_owner):
-                                    moved_to = spam_folder
-                                    logger.info(f"Auto-spam moved uid={uid.decode() if isinstance(uid, bytes) else str(uid)} to {spam_folder}: {spam_reason}")
+                                moved_to = ""
+                                if is_spam and auto_spam and spam_folder:
+                                    if _imap_move(uid, spam_folder, account_id=account_id, owner=account_owner):
+                                        moved_to = spam_folder
+                                        logger.info(f"Auto-spam moved uid={uid.decode() if isinstance(uid, bytes) else str(uid)} to {spam_folder}: {spam_reason}")
 
-                            _c = _sql3.connect(SCHEDULED_DB)
-                            _c.execute("""
-                                INSERT OR REPLACE INTO email_tags
-                                (message_id, owner, account_id, uid, folder, subject, sender, tags, spam_verdict,
-                                 spam_reason, moved_to, model_used, created_at)
-                                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-                            """, (message_id, account_owner or "", account_id or "", uid.decode() if isinstance(uid, bytes) else str(uid), _folder, subject, sender,
-                                  json.dumps(tags), 1 if is_spam else 0,
-                                  spam_reason, moved_to, model, datetime.utcnow().isoformat()))
-                            _c.commit()
-                            _c.close()
-                            _tag_existing.add(message_id)
+                                _c = _sql3.connect(SCHEDULED_DB)
+                                _c.execute("""
+                                    INSERT OR REPLACE INTO email_tags
+                                    (message_id, owner, uid, folder, subject, sender, tags, spam_verdict,
+                                     spam_reason, moved_to, model_used, created_at)
+                                    VALUES (?, ?, ?, 'INBOX', ?, ?, ?, ?, ?, ?, ?, ?)
+                                """, (message_id, account_owner or "", uid.decode() if isinstance(uid, bytes) else str(uid), subject, sender,
+                                      json.dumps(tags), 1 if is_spam else 0,
+                                      spam_reason, moved_to, model, datetime.utcnow().isoformat()))
+                                _c.commit()
+                                _c.close()
+                                _tag_existing.add(message_id)
                     except Exception as e:
                         logger.warning(f"Auto-classify {uid} failed: {e}")
 

routes/gallery/__init__.py

@@ -1,6 +0,0 @@
-"""Gallery route domain package (slice 2a, #4082/#4071).
-
-Contains gallery_routes.py and gallery_helpers.py, migrated from the flat
-routes/ directory. Backward-compat shims at routes/gallery_routes.py and
-routes/gallery_helpers.py re-export from here.
-"""

routes/gallery/gallery_helpers.py

@@ -1,145 +0,0 @@
-"""gallery_helpers.py — extracted helpers, models, and small utilities.
-
-Imported by gallery_routes.py."""
-
-"""Gallery routes — browsable library for photos and AI-generated images."""
-
-import logging
-from datetime import datetime
-from typing import Dict, Any, Optional
-
-from pydantic import BaseModel
-
-from core.database import GalleryImage
-from src.auth_helpers import _auth_disabled
-
-logger = logging.getLogger(__name__)
-
-
-# ---- Request schemas ----
-
-class GalleryPatch(BaseModel):
-    tags: Optional[str] = None
-    favorite: Optional[bool] = None
-    album_id: Optional[str] = None
-
-
-# ---- EXIF extraction ----
-
-def _extract_exif(content: bytes) -> dict:
-    """Extract EXIF metadata from image bytes. Returns dict of fields."""
-    result = {"width": None, "height": None}
-    try:
-        from PIL import Image
-        from io import BytesIO
-        img = Image.open(BytesIO(content))
-        # Read the raw EXIF before any transpose: exif_transpose strips the
-        # orientation tag and with it the parsed EXIF view.
-        exif = img._getexif() if hasattr(img, '_getexif') else None
-
-        # Record DISPLAY dimensions (EXIF-rotated), matching upload_handler.
-        # A phone photo with Orientation 6/8 is stored landscape but shown
-        # portrait, so the raw width/height swap the aspect ratio.
-        try:
-            from PIL import ImageOps
-            img = ImageOps.exif_transpose(img) or img
-        except Exception:
-            pass
-        result["width"] = img.width
-        result["height"] = img.height
-
-        if not exif:
-            return result
-
-        # EXIF tag IDs
-        # 271=Make, 272=Model, 306=DateTime, 36867=DateTimeOriginal
-        # 34853=GPSInfo
-        result["camera_make"] = str(exif.get(271, "")).strip() or None
-        result["camera_model"] = str(exif.get(272, "")).strip() or None
-
-        # Date taken
-        for tag_id in (36867, 36868, 306):  # DateTimeOriginal, DateTimeDigitized, DateTime
-            raw = exif.get(tag_id)
-            if raw:
-                try:
-                    result["taken_at"] = datetime.strptime(str(raw).strip(), "%Y:%m:%d %H:%M:%S")
-                    break
-                except (ValueError, TypeError):
-                    pass
-
-        # GPS
-        gps_info = exif.get(34853)
-        if gps_info and isinstance(gps_info, dict):
-            try:
-                def _to_deg(vals):
-                    d, m, s = [float(v) for v in vals]
-                    return d + m / 60 + s / 3600
-                if 2 in gps_info and 4 in gps_info:
-                    lat = _to_deg(gps_info[2])
-                    lng = _to_deg(gps_info[4])
-                    if gps_info.get(1) == 'S': lat = -lat
-                    if gps_info.get(3) == 'W': lng = -lng
-                    result["gps_lat"] = f"{lat:.6f}"
-                    result["gps_lng"] = f"{lng:.6f}"
-            except Exception:
-                pass
-    except Exception as e:
-        # User-visible failure (photo loses metadata): surface at WARNING
-        # and record on the result so the upload endpoint can pass it back.
-        logger.warning(f"EXIF extraction failed: {e}")
-        result["exif_error"] = str(e)
-    return result
-
-
-# ---- Helpers ----
-
-def _image_to_dict(img: GalleryImage, session_name: str = None) -> Dict[str, Any]:
-    return {
-        "id": img.id,
-        "filename": img.filename,
-        "url": f"/api/generated-image/{img.filename}",
-        "prompt": img.prompt,
-        "caption": img.caption or "",
-        "model": img.model,
-        "size": img.size,
-        "quality": img.quality,
-        "tags": img.tags or "",
-        "ai_tags": img.ai_tags or "",
-        "user_tags": img.tags or "",
-        "session_id": img.session_id,
-        "session_name": session_name,
-        "album_id": img.album_id,
-        "is_active": img.is_active,
-        "favorite": img.favorite or False,
-        "taken_at": img.taken_at.isoformat() if img.taken_at else None,
-        "camera": f"{img.camera_make or ''} {img.camera_model or ''}".strip() or None,
-        "gps": {"lat": img.gps_lat, "lng": img.gps_lng} if img.gps_lat else None,
-        "width": img.width,
-        "height": img.height,
-        "file_size": img.file_size,
-        "created_at": img.created_at.isoformat() if img.created_at else None,
-        "updated_at": img.updated_at.isoformat() if img.updated_at else None,
-    }
-
-
-def _owner_filter(q, user, model_cls=GalleryImage):
-    """Apply owner filtering to a gallery query.
-
-    ``get_current_user`` returns None both in auth-disabled single-user mode
-    and when auth is enabled but no current user was resolved. Preserve the
-    single-user behavior, but fail closed for auth-enabled null-user states.
-    """
-    if user is not None:
-        return q.filter(model_cls.owner == user)
-    if _auth_disabled():
-        return q
-    return q.filter(False)
-
-
-
-def _human_size(nbytes):
-    for unit in ['B', 'KB', 'MB', 'GB', 'TB']:
-        if abs(nbytes) < 1024:
-            return f"{nbytes:.1f} {unit}"
-        nbytes /= 1024
-    return f"{nbytes:.1f} PB"

routes/gallery_helpers.py

@@ -1,14 +1,144 @@
-"""Backward-compat shim - canonical location is routes/gallery/gallery_helpers.py.
+"""gallery_helpers.py — extracted helpers, models, and small utilities.
 
-This module is replaced in ``sys.modules`` by the canonical module object so
-that ``import routes.gallery_helpers``, ``from routes.gallery_helpers import X``,
-``importlib.import_module("routes.gallery_helpers")``, and
-``monkeypatch.setattr(routes.gallery_helpers, ...)`` all operate on the same
-object. Keeps existing import paths working after slice 2a (#4082/#4071).
-"""
+Imported by gallery_routes.py."""
 
-import sys as _sys
+"""Gallery routes — browsable library for photos and AI-generated images."""
 
-from routes.gallery import gallery_helpers as _canonical  # noqa: F401
+import logging
+from datetime import datetime
+from typing import Dict, Any, Optional
 
-_sys.modules[__name__] = _canonical
+from pydantic import BaseModel
+
+from core.database import GalleryImage
+from src.auth_helpers import _auth_disabled
+
+logger = logging.getLogger(__name__)
+
+
+# ---- Request schemas ----
+
+class GalleryPatch(BaseModel):
+    tags: Optional[str] = None
+    favorite: Optional[bool] = None
+    album_id: Optional[str] = None
+
+
+# ---- EXIF extraction ----
+
+def _extract_exif(content: bytes) -> dict:
+    """Extract EXIF metadata from image bytes. Returns dict of fields."""
+    result = {"width": None, "height": None}
+    try:
+        from PIL import Image
+        from io import BytesIO
+        img = Image.open(BytesIO(content))
+        # Read the raw EXIF before any transpose: exif_transpose strips the
+        # orientation tag and with it the parsed EXIF view.
+        exif = img._getexif() if hasattr(img, '_getexif') else None
+
+        # Record DISPLAY dimensions (EXIF-rotated), matching upload_handler.
+        # A phone photo with Orientation 6/8 is stored landscape but shown
+        # portrait, so the raw width/height swap the aspect ratio.
+        try:
+            from PIL import ImageOps
+            img = ImageOps.exif_transpose(img) or img
+        except Exception:
+            pass
+        result["width"] = img.width
+        result["height"] = img.height
+
+        if not exif:
+            return result
+
+        # EXIF tag IDs
+        # 271=Make, 272=Model, 306=DateTime, 36867=DateTimeOriginal
+        # 34853=GPSInfo
+        result["camera_make"] = str(exif.get(271, "")).strip() or None
+        result["camera_model"] = str(exif.get(272, "")).strip() or None
+
+        # Date taken
+        for tag_id in (36867, 36868, 306):  # DateTimeOriginal, DateTimeDigitized, DateTime
+            raw = exif.get(tag_id)
+            if raw:
+                try:
+                    result["taken_at"] = datetime.strptime(str(raw).strip(), "%Y:%m:%d %H:%M:%S")
+                    break
+                except (ValueError, TypeError):
+                    pass
+
+        # GPS
+        gps_info = exif.get(34853)
+        if gps_info and isinstance(gps_info, dict):
+            try:
+                def _to_deg(vals):
+                    d, m, s = [float(v) for v in vals]
+                    return d + m / 60 + s / 3600
+                if 2 in gps_info and 4 in gps_info:
+                    lat = _to_deg(gps_info[2])
+                    lng = _to_deg(gps_info[4])
+                    if gps_info.get(1) == 'S': lat = -lat
+                    if gps_info.get(3) == 'W': lng = -lng
+                    result["gps_lat"] = f"{lat:.6f}"
+                    result["gps_lng"] = f"{lng:.6f}"
+            except Exception:
+                pass
+    except Exception as e:
+        # User-visible failure (photo loses metadata): surface at WARNING
+        # and record on the result so the upload endpoint can pass it back.
+        logger.warning(f"EXIF extraction failed: {e}")
+        result["exif_error"] = str(e)
+    return result
+
+
+# ---- Helpers ----
+
+def _image_to_dict(img: GalleryImage, session_name: str = None) -> Dict[str, Any]:
+    return {
+        "id": img.id,
+        "filename": img.filename,
+        "url": f"/api/generated-image/{img.filename}",
+        "prompt": img.prompt,
+        "model": img.model,
+        "size": img.size,
+        "quality": img.quality,
+        "tags": img.tags or "",
+        "ai_tags": img.ai_tags or "",
+        "user_tags": img.tags or "",
+        "session_id": img.session_id,
+        "session_name": session_name,
+        "album_id": img.album_id,
+        "is_active": img.is_active,
+        "favorite": img.favorite or False,
+        "taken_at": img.taken_at.isoformat() if img.taken_at else None,
+        "camera": f"{img.camera_make or ''} {img.camera_model or ''}".strip() or None,
+        "gps": {"lat": img.gps_lat, "lng": img.gps_lng} if img.gps_lat else None,
+        "width": img.width,
+        "height": img.height,
+        "file_size": img.file_size,
+        "created_at": img.created_at.isoformat() if img.created_at else None,
+        "updated_at": img.updated_at.isoformat() if img.updated_at else None,
+    }
+
+
+def _owner_filter(q, user, model_cls=GalleryImage):
+    """Apply owner filtering to a gallery query.
+
+    ``get_current_user`` returns None both in auth-disabled single-user mode
+    and when auth is enabled but no current user was resolved. Preserve the
+    single-user behavior, but fail closed for auth-enabled null-user states.
+    """
+    if user is not None:
+        return q.filter(model_cls.owner == user)
+    if _auth_disabled():
+        return q
+    return q.filter(False)
+
+
+
+def _human_size(nbytes):
+    for unit in ['B', 'KB', 'MB', 'GB', 'TB']:
+        if abs(nbytes) < 1024:
+            return f"{nbytes:.1f} {unit}"
+        nbytes /= 1024
+    return f"{nbytes:.1f} PB"

routes/history/__init__.py

@@ -1,5 +0,0 @@
-"""History route domain package (slice 2d, #4082/#4071).
-
-Contains history_routes.py, migrated from the flat routes/ directory.
-Backward-compat shim at routes/history_routes.py re-exports from here.
-"""

routes/history/history_routes.py

@@ -1,768 +0,0 @@
-"""History routes — session history, truncation, fork, conversation topics."""
-
-import json
-import uuid
-import logging
-import re
-from typing import Dict, Any, Optional
-
-from fastapi import APIRouter, Request, HTTPException
-
-from core.models import ChatMessage
-from core.database import SessionLocal, ChatMessage as DbChatMessage, Session as DbSession
-from src.topic_analyzer import analyze_topics
-from routes.session_routes import (
-    _message_role,
-    _message_text,
-    _reject_compact_during_active_run,
-    _verify_session_owner,
-)
-
-logger = logging.getLogger(__name__)
-
-_HISTORY_INLINE_MEDIA_THRESHOLD = 200_000
-_DATA_IMAGE_RE = re.compile(r"data:image/[^;,\"]+;base64,[A-Za-z0-9+/=\s]+")
-
-
-def _history_display_content(content: Any) -> Any:
-    """Return a lightweight browser-display copy of stored message content.
-
-    Older multimodal user messages may be persisted as a JSON *string*
-    containing image_url blocks with inline base64 image bytes. Those bytes are
-    needed for model calls when the turn is first sent, but they should not be
-    sent back through /api/history every time the user opens the chat. The
-    attachment metadata already carries file ids/names for the UI cards.
-    """
-    if isinstance(content, list):
-        text_parts = []
-        omitted_media = 0
-        for block in content:
-            if not isinstance(block, dict):
-                continue
-            if block.get("type") == "text":
-                text = block.get("text")
-                if isinstance(text, str) and text:
-                    text_parts.append(text)
-            elif block.get("type") in {"image_url", "input_image", "audio", "input_audio"}:
-                omitted_media += 1
-        text = "\n".join(text_parts).strip()
-        if omitted_media and not text:
-            return f"[{omitted_media} media attachment{'s' if omitted_media != 1 else ''} omitted from history view]"
-        return text
-
-    if not isinstance(content, str):
-        return content
-    if len(content) < _HISTORY_INLINE_MEDIA_THRESHOLD and "data:image/" not in content:
-        return content
-
-    stripped = content.lstrip()
-    if stripped.startswith("["):
-        try:
-            blocks = json.loads(content)
-        except (json.JSONDecodeError, TypeError, ValueError):
-            blocks = None
-        if isinstance(blocks, list):
-            text_parts = []
-            for block in blocks:
-                if not isinstance(block, dict):
-                    continue
-                if block.get("type") == "text":
-                    text = block.get("text")
-                    if isinstance(text, str) and text:
-                        text_parts.append(text)
-            if text_parts:
-                return "\n".join(text_parts).strip()
-
-    if "data:image/" in content:
-        return _DATA_IMAGE_RE.sub("[inline image omitted from history view]", content)
-    return content
-
-
-def _merge_continue_rows_to_delete(db_messages, db1, db2):
-    """DB rows to delete when merging the last two assistant messages.
-
-    Always the second assistant message (db2), plus ONLY the single
-    intervening "continue" user message (the one carrying "previous response
-    was interrupted") — matching the in-memory merge. The previous code
-    deleted the whole index range between the two assistant rows, destroying
-    any tool/system/user messages in between and desyncing the DB from the
-    in-memory history.
-    """
-    to_delete = [db2]
-    i1 = next((i for i, m in enumerate(db_messages) if m is db1), None)
-    i2 = next((i for i, m in enumerate(db_messages) if m is db2), None)
-    if i1 is not None and i2 is not None and i2 - 1 > i1:
-        between = db_messages[i2 - 1]
-        if getattr(between, "role", "") == "user" and            "previous response was interrupted" in (getattr(between, "content", "") or ""):
-            to_delete.append(between)
-    return to_delete
-
-
-def setup_history_routes(session_manager) -> APIRouter:
-    router = APIRouter(tags=["history"])
-
-    def _db_history_entry(m: DbChatMessage) -> Dict[str, Any]:
-        entry = {"role": m.role, "content": _history_display_content(m.content)}
-        meta = {}
-        if m.meta_data:
-            try:
-                meta = json.loads(m.meta_data) or {}
-            except (json.JSONDecodeError, ValueError):
-                meta = {}
-        if m.timestamp and "timestamp" not in meta:
-            meta["timestamp"] = m.timestamp.isoformat() + "Z"
-        if meta:
-            entry["metadata"] = meta
-        return entry
-
-    @router.get("/api/history/{session_id}")
-    async def get_session_history(
-        request: Request,
-        session_id: str,
-        limit: Optional[int] = None,
-        offset: Optional[int] = None,
-    ) -> Dict[str, Any]:
-        _verify_session_owner(request, session_id)
-        if limit is not None:
-            page_limit = max(1, min(int(limit), 100))
-            db = SessionLocal()
-            try:
-                db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
-                if db_session is None:
-                    raise HTTPException(404, f"Session '{session_id}' not found")
-
-                total = (
-                    db.query(DbChatMessage)
-                    .filter(DbChatMessage.session_id == session_id)
-                    .count()
-                )
-                page_offset = int(offset) if offset is not None else max(total - page_limit, 0)
-                page_offset = max(0, min(page_offset, total))
-                rows = (
-                    db.query(DbChatMessage)
-                    .filter(DbChatMessage.session_id == session_id)
-                    .order_by(DbChatMessage.timestamp)
-                    .offset(page_offset)
-                    .limit(page_limit)
-                    .all()
-                )
-                history_dict = [
-                    entry for entry in (_db_history_entry(m) for m in rows)
-                    if not (entry.get("metadata") or {}).get("hidden")
-                ]
-                return {
-                    "history": history_dict,
-                    "model": db_session.model,
-                    "endpoint_url": db_session.endpoint_url,
-                    "name": db_session.name,
-                    "offset": page_offset,
-                    "limit": page_limit,
-                    "total": total,
-                    "has_more_before": page_offset > 0,
-                    "has_more_after": page_offset + len(rows) < total,
-                }
-            finally:
-                db.close()
-
-        try:
-            session = session_manager.get_session(session_id)
-        except KeyError:
-            raise HTTPException(404, f"Session '{session_id}' not found")
-
-        history_dict = []
-        for msg in session.history:
-            if isinstance(msg, ChatMessage):
-                # Skip hidden messages (e.g. compaction summaries for AI context)
-                if msg.metadata and msg.metadata.get("hidden"):
-                    continue
-                entry = {"role": msg.role, "content": _history_display_content(msg.content)}
-                if msg.metadata:
-                    entry["metadata"] = msg.metadata
-                history_dict.append(entry)
-            elif isinstance(msg, dict):
-                if msg.get("metadata", {}).get("hidden"):
-                    continue
-                entry = {
-                    "role": msg.get("role", ""),
-                    "content": _history_display_content(msg.get("content", "")),
-                }
-                if msg.get("metadata"):
-                    entry["metadata"] = msg["metadata"]
-                history_dict.append(entry)
-
-        # Fallback: load from DB if in-memory is empty
-        if not history_dict:
-            db = SessionLocal()
-            try:
-                db_messages = (
-                    db.query(DbChatMessage)
-                    .filter(DbChatMessage.session_id == session_id)
-                    .order_by(DbChatMessage.timestamp)
-                    .all()
-                )
-                db_history = []
-                for m in db_messages:
-                    db_history.append(_db_history_entry(m))
-                if db_history:
-                    # Rebuild in-memory history from the full set so hidden
-                    # messages (e.g. compaction summaries) are kept for AI context.
-                    session.history = [
-                        ChatMessage(role=m["role"], content=m["content"], metadata=m.get("metadata"))
-                        for m in db_history
-                    ]
-                # Response excludes hidden messages, matching the in-memory path.
-                history_dict = [
-                    m for m in db_history
-                    if not (m.get("metadata") or {}).get("hidden")
-                ]
-            except Exception as e:
-                logger.error(f"DB fallback failed for {session_id}: {e}")
-            finally:
-                db.close()
-
-        return {
-            "history": history_dict,
-            "model": session.model,
-            "endpoint_url": session.endpoint_url,
-            "name": session.name,
-        }
-
-    @router.post("/api/session/{session_id}/truncate")
-    async def truncate_session(request: Request, session_id: str):
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            keep_count = body.get("keep_count", 0)
-            result = session_manager.truncate_messages(session_id, keep_count)
-            return {"status": "ok", "kept": keep_count, "truncated": result}
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        except Exception as e:
-            logger.error(f"Truncate error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.post("/api/session/{session_id}/message")
-    async def add_message(request: Request, session_id: str):
-        """Add a message to a session (for slash command persistence)."""
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            role = body.get("role", "assistant")
-            content = body.get("content", "")
-            if not content:
-                raise HTTPException(400, "content is required")
-            msg = ChatMessage(role=role, content=content, metadata=body.get("metadata"))
-            session_manager.add_message(session_id, msg)
-            return {"status": "ok"}
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-
-    @router.post("/api/session/{session_id}/delete-messages")
-    async def delete_messages(request: Request, session_id: str):
-        """Delete specific messages by DB ID (or legacy index)."""
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            msg_ids = body.get("msg_ids", [])
-            indices = body.get("indices")  # legacy fallback
-
-            session = session_manager.get_session(session_id)
-            db = SessionLocal()
-            try:
-                if msg_ids:
-                    # New ID-based delete
-                    deleted = 0
-                    for mid in msg_ids:
-                        db_msg = db.query(DbChatMessage).filter(
-                            DbChatMessage.id == mid,
-                            DbChatMessage.session_id == session_id,
-                        ).first()
-                        if db_msg:
-                            db.delete(db_msg)
-                            deleted += 1
-
-                    # Remove from in-memory history by matching _db_id
-                    def _get_db_id(m):
-                        meta = m.metadata if isinstance(m, ChatMessage) else (m.get('metadata') if isinstance(m, dict) else None)
-                        return meta.get('_db_id') if isinstance(meta, dict) else None
-                    session.history = [m for m in session.history if _get_db_id(m) not in msg_ids]
-                elif indices:
-                    # Legacy index-based delete
-                    indices = sorted(indices, reverse=True)
-                    db_messages = db.query(DbChatMessage).filter(
-                        DbChatMessage.session_id == session_id
-                    ).order_by(DbChatMessage.timestamp).all()
-
-                    deleted = 0
-                    for idx in indices:
-                        if 0 <= idx < len(db_messages):
-                            db.delete(db_messages[idx])
-                            deleted += 1
-                        if 0 <= idx < len(session.history):
-                            session.history.pop(idx)
-                else:
-                    return {"status": "ok", "deleted": 0}
-
-                session.message_count = len(session.history)
-                db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
-                if db_session:
-                    db_session.message_count = len(session.history)
-                    from datetime import datetime, timezone
-                    db_session.updated_at = datetime.now(timezone.utc)
-
-                db.commit()
-                return {"status": "ok", "deleted": deleted}
-            finally:
-                db.close()
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        except Exception as e:
-            logger.error(f"Delete messages error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.post("/api/session/{session_id}/edit-message")
-    async def edit_message(request: Request, session_id: str):
-        """Edit the content of a message by its database ID."""
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            msg_id = body.get("msg_id")
-            content = body.get("content")
-            if not msg_id or content is None:
-                raise HTTPException(400, "msg_id and content are required")
-
-            session = session_manager.get_session(session_id)
-            db = SessionLocal()
-            try:
-                db_msg = db.query(DbChatMessage).filter(
-                    DbChatMessage.id == msg_id,
-                    DbChatMessage.session_id == session_id,
-                ).first()
-                if not db_msg:
-                    raise HTTPException(404, "Message not found")
-
-                db_msg.content = content
-                meta = {}
-                if db_msg.meta_data:
-                    try: meta = json.loads(db_msg.meta_data)
-                    except (json.JSONDecodeError, ValueError): pass
-                meta['edited'] = True
-                db_msg.meta_data = json.dumps(meta)
-
-                # Update in-memory history by matching _db_id
-                for hmsg in session.history:
-                    hmeta = hmsg.metadata if isinstance(hmsg, ChatMessage) else hmsg.get('metadata')
-                    if isinstance(hmeta, dict) and hmeta.get('_db_id') == msg_id:
-                        if isinstance(hmsg, ChatMessage):
-                            hmsg.content = content
-                            hmsg.metadata['edited'] = True
-                        elif isinstance(hmsg, dict):
-                            hmsg['content'] = content
-                            hmsg['metadata']['edited'] = True
-                        break
-
-                db.commit()
-                return {"status": "ok"}
-            finally:
-                db.close()
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        except HTTPException:
-            raise
-        except Exception as e:
-            logger.error(f"Edit message error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.post("/api/session/{session_id}/mark-stopped")
-    async def mark_stopped(request: Request, session_id: str):
-        """Mark the last assistant message as stopped by user."""
-        _verify_session_owner(request, session_id)
-        try:
-            session = session_manager.get_session(session_id)
-            # Find last assistant message and add stopped metadata
-            for msg in reversed(session.history):
-                if (isinstance(msg, ChatMessage) and msg.role == 'assistant') or \
-                   (isinstance(msg, dict) and msg.get('role') == 'assistant'):
-                    if isinstance(msg, ChatMessage):
-                        if not msg.metadata:
-                            msg.metadata = {}
-                        msg.metadata['stopped'] = True
-                        if not msg.metadata.get('model'):
-                            msg.metadata['model'] = session.model
-                    else:
-                        if 'metadata' not in msg:
-                            msg['metadata'] = {}
-                        msg['metadata']['stopped'] = True
-                        if not msg['metadata'].get('model'):
-                            msg['metadata']['model'] = session.model
-                    break
-            # Also update in DB
-            db = SessionLocal()
-            try:
-                import json as _json
-                db_messages = (
-                    db.query(DbChatMessage)
-                    .filter(DbChatMessage.session_id == session_id, DbChatMessage.role == 'assistant')
-                    .order_by(DbChatMessage.timestamp.desc())
-                    .first()
-                )
-                if db_messages:
-                    meta = {}
-                    if db_messages.meta_data:
-                        try:
-                            meta = _json.loads(db_messages.meta_data)
-                        except (json.JSONDecodeError, ValueError):
-                            pass
-                    meta['stopped'] = True
-                    if not meta.get('model'):
-                        meta['model'] = session.model
-                    db_messages.meta_data = _json.dumps(meta)
-                    db.commit()
-            finally:
-                db.close()
-            session_manager.save_sessions()
-            return {"status": "ok"}
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        except Exception as e:
-            logger.error(f"Mark stopped error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.post("/api/session/{session_id}/update-last-meta")
-    async def update_last_meta(request: Request, session_id: str):
-        """Merge metadata into the last assistant message (e.g. save variants)."""
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            meta_update = body.get("metadata", {})
-            session = session_manager.get_session(session_id)
-
-            # Update in-memory
-            for msg in reversed(session.history):
-                if (isinstance(msg, ChatMessage) and msg.role == 'assistant') or \
-                   (isinstance(msg, dict) and msg.get('role') == 'assistant'):
-                    if isinstance(msg, ChatMessage):
-                        if not msg.metadata:
-                            msg.metadata = {}
-                        msg.metadata.update(meta_update)
-                    else:
-                        if 'metadata' not in msg:
-                            msg['metadata'] = {}
-                        msg['metadata'].update(meta_update)
-                    break
-
-            # Update in DB
-            db = SessionLocal()
-            try:
-                import json as _json
-                db_msg = (
-                    db.query(DbChatMessage)
-                    .filter(DbChatMessage.session_id == session_id, DbChatMessage.role == 'assistant')
-                    .order_by(DbChatMessage.timestamp.desc())
-                    .first()
-                )
-                if db_msg:
-                    meta = {}
-                    if db_msg.meta_data:
-                        try: meta = _json.loads(db_msg.meta_data)
-                        except (json.JSONDecodeError, ValueError): pass
-                    meta.update(meta_update)
-                    db_msg.meta_data = _json.dumps(meta)
-                    db.commit()
-            finally:
-                db.close()
-            session_manager.save_sessions()
-            return {"status": "ok"}
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        except Exception as e:
-            logger.error(f"Update last meta error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.post("/api/session/{session_id}/merge-last-assistant")
-    async def merge_last_assistant(request: Request, session_id: str):
-        """Merge the last two assistant messages into one (for continue)."""
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            separator = body.get("separator", "\n\n")
-            session = session_manager.get_session(session_id)
-
-            # Find last two assistant messages in-memory
-            ai_indices = []
-            for i, msg in enumerate(session.history):
-                role = msg.role if isinstance(msg, ChatMessage) else msg.get('role', '')
-                if role == 'assistant':
-                    ai_indices.append(i)
-
-            if len(ai_indices) < 2:
-                return {"status": "ok", "merged": False}
-
-            idx1, idx2 = ai_indices[-2], ai_indices[-1]
-            msg1, msg2 = session.history[idx1], session.history[idx2]
-
-            content1 = msg1.content if isinstance(msg1, ChatMessage) else msg1.get('content', '')
-            content2 = msg2.content if isinstance(msg2, ChatMessage) else msg2.get('content', '')
-            merged_content = content1 + separator + content2
-
-            # Merge metadata
-            meta1 = (msg1.metadata if isinstance(msg1, ChatMessage) else msg1.get('metadata')) or {}
-            meta2 = (msg2.metadata if isinstance(msg2, ChatMessage) else msg2.get('metadata')) or {}
-            merged_meta = {**meta1, **meta2}
-            merged_meta.pop('stopped', None)  # no longer stopped after continue
-
-            # Update first message, remove second
-            if isinstance(msg1, ChatMessage):
-                msg1.content = merged_content
-                msg1.metadata = merged_meta
-            else:
-                msg1['content'] = merged_content
-                msg1['metadata'] = merged_meta
-
-            # Also remove the hidden "continue" user message between them if present
-            # It's the message at idx2-1 if it's a user message with continue text
-            remove_indices = [idx2]
-            if idx2 - 1 > idx1:
-                between = session.history[idx2 - 1]
-                between_role = between.role if isinstance(between, ChatMessage) else between.get('role', '')
-                between_content = between.content if isinstance(between, ChatMessage) else between.get('content', '')
-                if between_role == 'user' and 'previous response was interrupted' in between_content:
-                    remove_indices.insert(0, idx2 - 1)
-
-            for ri in sorted(remove_indices, reverse=True):
-                session.history.pop(ri)
-
-            # Update DB
-            db = SessionLocal()
-            try:
-                import json as _json
-                db_messages = (
-                    db.query(DbChatMessage)
-                    .filter(DbChatMessage.session_id == session_id)
-                    .order_by(DbChatMessage.timestamp)
-                    .all()
-                )
-                # Find last two assistant messages in DB
-                ai_db = [(i, m) for i, m in enumerate(db_messages) if m.role == 'assistant']
-                if len(ai_db) >= 2:
-                    (_, db1), (_, db2) = ai_db[-2], ai_db[-1]
-                    db1.content = merged_content
-                    db1.meta_data = _json.dumps(merged_meta)
-
-                    # Mirror the in-memory deletion: remove the second assistant
-                    # message and ONLY the "continue" user message between them
-                    # (not arbitrary tool/system/user rows). The old
-                    # range-delete destroyed every row between the two assistant
-                    # messages, desyncing the DB from the in-memory history.
-                    for _row in _merge_continue_rows_to_delete(db_messages, db1, db2):
-                        db.delete(_row)
-
-                    db.commit()
-            finally:
-                db.close()
-            session_manager.save_sessions()
-            return {"status": "ok", "merged": True}
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        except Exception as e:
-            logger.error(f"Merge assistant error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.post("/api/session/{session_id}/fork")
-    async def fork_session(request: Request, session_id: str):
-        """Create a new session with messages copied up to keep_count."""
-        _verify_session_owner(request, session_id)
-        try:
-            body = await request.json()
-            keep_count = body.get("keep_count", 0)
-
-            # Get the source session
-            source = session_manager.sessions.get(session_id)
-            if not source:
-                raise HTTPException(404, "Session not found")
-
-            # Create new session
-            new_id = str(uuid.uuid4())
-            fork_name = f"\u2ADD {source.name}"
-            new_session = session_manager.create_session(
-                session_id=new_id,
-                name=fork_name,
-                endpoint_url=source.endpoint_url,
-                model=source.model,
-                rag=False,
-                owner=getattr(source, 'owner', None),
-            )
-
-            # Copy messages up to keep_count
-            msgs_to_copy = source.history[:keep_count]
-            for msg in msgs_to_copy:
-                # Copy the metadata dict. Sharing it would let the fork's
-                # persistence (add_message -> _persist_message stamps
-                # _db_id/timestamp onto the dict) mutate the SOURCE session's
-                # in-memory messages, corrupting their _db_id and breaking
-                # edit/delete-by-id on the original conversation.
-                meta = dict(msg.metadata) if isinstance(msg.metadata, dict) else None
-                new_session.add_message(ChatMessage(msg.role, msg.content, meta))
-            try:
-                from src.event_bus import fire_event
-                fire_event("session_created", getattr(source, 'owner', None))
-            except Exception:
-                logger.debug("session_created event dispatch failed", exc_info=True)
-
-            return {
-                "status": "ok",
-                "id": new_id,
-                "name": fork_name,
-                "kept": len(msgs_to_copy),
-            }
-        except HTTPException:
-            raise
-        except Exception as e:
-            logger.error(f"Fork error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    @router.get("/api/conversations/topics")
-    async def get_conversation_topics(request: Request) -> Dict[str, Any]:
-        from src.auth_helpers import require_user
-        user = require_user(request)
-        try:
-            return analyze_topics(session_manager, owner=user or None)
-        except Exception as e:
-            raise HTTPException(500, f"Topic analysis failed: {e}")
-
-    @router.post("/api/session/{session_id}/compact")
-    async def compact_session(request: Request, session_id: str):
-        """Manually trigger context compaction for a session."""
-        _verify_session_owner(request, session_id)
-        from src.auth_helpers import effective_user
-        owner = effective_user(request)
-        try:
-            session = session_manager.get_session(session_id)
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        _reject_compact_during_active_run(session_id)
-
-        try:
-            from src.model_context import estimate_tokens, get_context_length
-            from src.llm_core import llm_call_async
-            from src.endpoint_resolver import resolve_endpoint
-
-            if len(session.history) < 6:
-                return {"status": "ok", "message": "Not enough messages to compact"}
-
-            ctx_len = get_context_length(session.endpoint_url, session.model)
-            messages_before = session.get_context_messages()
-            used_before = estimate_tokens(messages_before)
-            pct_before = round((used_before / ctx_len) * 100, 1) if ctx_len else 0
-            msg_count_before = len(session.history)
-
-            # Keep only last 4 messages, summarize the rest
-            keep_count = 4
-            older = session.history[:-keep_count]
-            recent = session.history[-keep_count:]
-
-            # Build text to summarize
-            convo_text = "\n".join(
-                f"{_message_role(m).upper()}: "
-                f"{_message_text(m)[:2000]}"
-                for m in older
-            )
-
-            # Use utility model if available
-            util_url, util_model, util_headers = resolve_endpoint("utility", owner=owner or None)
-            compact_url = util_url or session.endpoint_url
-            compact_model = util_model or session.model
-            compact_headers = util_headers if util_url else session.headers
-
-            from src.context_compactor import SELF_SUMMARY_SYSTEM_PROMPT
-            compaction_count = sum(1 for m in session.history if isinstance(m, ChatMessage) and "[Conversation summary" in (m.content or ""))
-            sys_prompt = SELF_SUMMARY_SYSTEM_PROMPT.replace("{count}", str(len(older))).replace("{n}", str(compaction_count + 1))
-            summary = await llm_call_async(
-                compact_url, compact_model,
-                [
-                    {"role": "system", "content": sys_prompt},
-                    {"role": "user", "content": convo_text},
-                ],
-                temperature=0.2, max_tokens=1024,
-                headers=compact_headers, timeout=30,
-            )
-
-            # Replace session history: summary as system message + recent messages
-            # System message holds the full summary for AI context
-            system_summary = ChatMessage(
-                role="system",
-                content=f"[Conversation summary — {len(older)} earlier messages were compacted]\n\n{summary}",
-                metadata={"compacted": True, "hidden": True},
-            )
-            # Visible assistant message just shows stats
-            summary_msg = ChatMessage(
-                role="assistant",
-                content=f"**Conversation compacted** — {len(older)} messages summarized, {len(recent)} kept.",
-                metadata={"compacted": True, "messages_removed": len(older)},
-            )
-            new_history = [system_summary, summary_msg] + list(recent)
-            session.history = new_history
-            session.message_count = len(session.history)
-            logger.info(f"Compact: session {session_id} history now has {len(session.history)} messages (was {msg_count_before})")
-
-            # Update DB: delete old messages, insert summary
-            db = SessionLocal()
-            try:
-                db_msgs = db.query(DbChatMessage).filter(
-                    DbChatMessage.session_id == session_id
-                ).order_by(DbChatMessage.timestamp).all()
-
-                # Delete all but the last keep_count
-                for m in db_msgs[:-keep_count]:
-                    db.delete(m)
-
-                # Insert system summary (hidden, for AI context) and visible summary
-                import json as _json
-                import uuid
-                from datetime import datetime, timezone
-                now = datetime.now(timezone.utc)
-                db_sys_summary = DbChatMessage(
-                    id=str(uuid.uuid4()),
-                    session_id=session_id,
-                    role="system",
-                    content=system_summary.content,
-                    meta_data=_json.dumps(system_summary.metadata),
-                    timestamp=now,
-                )
-                db.add(db_sys_summary)
-                db_summary = DbChatMessage(
-                    id=str(uuid.uuid4()),
-                    session_id=session_id,
-                    role="assistant",
-                    content=summary_msg.content,
-                    meta_data=_json.dumps(summary_msg.metadata),
-                    timestamp=now,
-                )
-                db.add(db_summary)
-
-                # Update session record
-                db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
-                if db_session:
-                    db_session.message_count = len(session.history)
-                    db_session.updated_at = datetime.now(timezone.utc)
-                db.commit()
-            finally:
-                db.close()
-
-            session_manager.save_sessions()
-
-            used_after = estimate_tokens(session.get_context_messages())
-            pct_after = round((used_after / ctx_len) * 100, 1) if ctx_len else 0
-
-            return {
-                "status": "ok",
-                "message": f"Compacted: {msg_count_before} msgs → {len(session.history)} msgs ({pct_before}% → {pct_after}%)",
-                "before": pct_before,
-                "after": pct_after,
-            }
-
-        except Exception as e:
-            logger.error(f"Manual compact error {session_id}: {e}")
-            raise HTTPException(500, str(e))
-
-    return router

routes/history_routes.py

@@ -1,17 +1,662 @@
-"""Backward-compat shim — canonical location is routes/history/history_routes.py.
+"""History routes — session history, truncation, fork, conversation topics."""
 
-This module is replaced in ``sys.modules`` by the canonical module object so
-that ``import routes.history_routes``, ``from routes.history_routes import X``,
-``importlib.import_module("routes.history_routes")``, and the
-``import ... as history_routes`` + ``monkeypatch.setattr(history_routes, ...)``
-pattern used by test_history_compact_tool_calls.py / test_fork_session_metadata.py
-all operate on the *same* object the application actually uses. Keeps existing
-import paths working after slice 2d (#4082/#4071). Source-introspection tests
-read the canonical file by path.
-"""
+import json
+import uuid
+import logging
+from typing import Dict, Any
 
-import sys as _sys
+from fastapi import APIRouter, Request, HTTPException
 
-from routes.history import history_routes as _canonical  # noqa: F401
+from core.models import ChatMessage
+from core.database import SessionLocal, ChatMessage as DbChatMessage, Session as DbSession
+from src.topic_analyzer import analyze_topics
+from routes.session_routes import (
+    _message_role,
+    _message_text,
+    _reject_compact_during_active_run,
+    _verify_session_owner,
+)
 
-_sys.modules[__name__] = _canonical
+logger = logging.getLogger(__name__)
+
+
+def _merge_continue_rows_to_delete(db_messages, db1, db2):
+    """DB rows to delete when merging the last two assistant messages.
+
+    Always the second assistant message (db2), plus ONLY the single
+    intervening "continue" user message (the one carrying "previous response
+    was interrupted") — matching the in-memory merge. The previous code
+    deleted the whole index range between the two assistant rows, destroying
+    any tool/system/user messages in between and desyncing the DB from the
+    in-memory history.
+    """
+    to_delete = [db2]
+    i1 = next((i for i, m in enumerate(db_messages) if m is db1), None)
+    i2 = next((i for i, m in enumerate(db_messages) if m is db2), None)
+    if i1 is not None and i2 is not None and i2 - 1 > i1:
+        between = db_messages[i2 - 1]
+        if getattr(between, "role", "") == "user" and            "previous response was interrupted" in (getattr(between, "content", "") or ""):
+            to_delete.append(between)
+    return to_delete
+
+
+def setup_history_routes(session_manager) -> APIRouter:
+    router = APIRouter(tags=["history"])
+
+    @router.get("/api/history/{session_id}")
+    async def get_session_history(request: Request, session_id: str) -> Dict[str, Any]:
+        _verify_session_owner(request, session_id)
+        try:
+            session = session_manager.get_session(session_id)
+        except KeyError:
+            raise HTTPException(404, f"Session '{session_id}' not found")
+
+        history_dict = []
+        for msg in session.history:
+            if isinstance(msg, ChatMessage):
+                # Skip hidden messages (e.g. compaction summaries for AI context)
+                if msg.metadata and msg.metadata.get("hidden"):
+                    continue
+                entry = {"role": msg.role, "content": msg.content}
+                if msg.metadata:
+                    entry["metadata"] = msg.metadata
+                history_dict.append(entry)
+            elif isinstance(msg, dict):
+                if msg.get("metadata", {}).get("hidden"):
+                    continue
+                entry = {
+                    "role": msg.get("role", ""),
+                    "content": msg.get("content", ""),
+                }
+                if msg.get("metadata"):
+                    entry["metadata"] = msg["metadata"]
+                history_dict.append(entry)
+
+        # Fallback: load from DB if in-memory is empty
+        if not history_dict:
+            db = SessionLocal()
+            try:
+                db_messages = (
+                    db.query(DbChatMessage)
+                    .filter(DbChatMessage.session_id == session_id)
+                    .order_by(DbChatMessage.timestamp)
+                    .all()
+                )
+                import json as _json
+                db_history = []
+                for m in db_messages:
+                    entry = {"role": m.role, "content": m.content}
+                    meta = {}
+                    if m.meta_data:
+                        try:
+                            meta = _json.loads(m.meta_data) or {}
+                        except (json.JSONDecodeError, ValueError):
+                            meta = {}
+                    if m.timestamp and "timestamp" not in meta:
+                        meta["timestamp"] = m.timestamp.isoformat() + "Z"
+                    if meta:
+                        entry["metadata"] = meta
+                    db_history.append(entry)
+                if db_history:
+                    # Rebuild in-memory history from the full set so hidden
+                    # messages (e.g. compaction summaries) are kept for AI context.
+                    session.history = [
+                        ChatMessage(role=m["role"], content=m["content"], metadata=m.get("metadata"))
+                        for m in db_history
+                    ]
+                # Response excludes hidden messages, matching the in-memory path.
+                history_dict = [
+                    m for m in db_history
+                    if not (m.get("metadata") or {}).get("hidden")
+                ]
+            except Exception as e:
+                logger.error(f"DB fallback failed for {session_id}: {e}")
+            finally:
+                db.close()
+
+        return {
+            "history": history_dict,
+            "model": session.model,
+            "endpoint_url": session.endpoint_url,
+            "name": session.name,
+        }
+
+    @router.post("/api/session/{session_id}/truncate")
+    async def truncate_session(request: Request, session_id: str):
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            keep_count = body.get("keep_count", 0)
+            result = session_manager.truncate_messages(session_id, keep_count)
+            return {"status": "ok", "kept": keep_count, "truncated": result}
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        except Exception as e:
+            logger.error(f"Truncate error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.post("/api/session/{session_id}/message")
+    async def add_message(request: Request, session_id: str):
+        """Add a message to a session (for slash command persistence)."""
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            role = body.get("role", "assistant")
+            content = body.get("content", "")
+            if not content:
+                raise HTTPException(400, "content is required")
+            msg = ChatMessage(role=role, content=content, metadata=body.get("metadata"))
+            session_manager.add_message(session_id, msg)
+            return {"status": "ok"}
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+
+    @router.post("/api/session/{session_id}/delete-messages")
+    async def delete_messages(request: Request, session_id: str):
+        """Delete specific messages by DB ID (or legacy index)."""
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            msg_ids = body.get("msg_ids", [])
+            indices = body.get("indices")  # legacy fallback
+
+            session = session_manager.get_session(session_id)
+            db = SessionLocal()
+            try:
+                if msg_ids:
+                    # New ID-based delete
+                    deleted = 0
+                    for mid in msg_ids:
+                        db_msg = db.query(DbChatMessage).filter(
+                            DbChatMessage.id == mid,
+                            DbChatMessage.session_id == session_id,
+                        ).first()
+                        if db_msg:
+                            db.delete(db_msg)
+                            deleted += 1
+
+                    # Remove from in-memory history by matching _db_id
+                    def _get_db_id(m):
+                        meta = m.metadata if isinstance(m, ChatMessage) else (m.get('metadata') if isinstance(m, dict) else None)
+                        return meta.get('_db_id') if isinstance(meta, dict) else None
+                    session.history = [m for m in session.history if _get_db_id(m) not in msg_ids]
+                elif indices:
+                    # Legacy index-based delete
+                    indices = sorted(indices, reverse=True)
+                    db_messages = db.query(DbChatMessage).filter(
+                        DbChatMessage.session_id == session_id
+                    ).order_by(DbChatMessage.timestamp).all()
+
+                    deleted = 0
+                    for idx in indices:
+                        if 0 <= idx < len(db_messages):
+                            db.delete(db_messages[idx])
+                            deleted += 1
+                        if 0 <= idx < len(session.history):
+                            session.history.pop(idx)
+                else:
+                    return {"status": "ok", "deleted": 0}
+
+                session.message_count = len(session.history)
+                db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
+                if db_session:
+                    db_session.message_count = len(session.history)
+                    from datetime import datetime, timezone
+                    db_session.updated_at = datetime.now(timezone.utc)
+
+                db.commit()
+                return {"status": "ok", "deleted": deleted}
+            finally:
+                db.close()
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        except Exception as e:
+            logger.error(f"Delete messages error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.post("/api/session/{session_id}/edit-message")
+    async def edit_message(request: Request, session_id: str):
+        """Edit the content of a message by its database ID."""
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            msg_id = body.get("msg_id")
+            content = body.get("content")
+            if not msg_id or content is None:
+                raise HTTPException(400, "msg_id and content are required")
+
+            session = session_manager.get_session(session_id)
+            db = SessionLocal()
+            try:
+                db_msg = db.query(DbChatMessage).filter(
+                    DbChatMessage.id == msg_id,
+                    DbChatMessage.session_id == session_id,
+                ).first()
+                if not db_msg:
+                    raise HTTPException(404, "Message not found")
+
+                db_msg.content = content
+                meta = {}
+                if db_msg.meta_data:
+                    try: meta = json.loads(db_msg.meta_data)
+                    except (json.JSONDecodeError, ValueError): pass
+                meta['edited'] = True
+                db_msg.meta_data = json.dumps(meta)
+
+                # Update in-memory history by matching _db_id
+                for hmsg in session.history:
+                    hmeta = hmsg.metadata if isinstance(hmsg, ChatMessage) else hmsg.get('metadata')
+                    if isinstance(hmeta, dict) and hmeta.get('_db_id') == msg_id:
+                        if isinstance(hmsg, ChatMessage):
+                            hmsg.content = content
+                            hmsg.metadata['edited'] = True
+                        elif isinstance(hmsg, dict):
+                            hmsg['content'] = content
+                            hmsg['metadata']['edited'] = True
+                        break
+
+                db.commit()
+                return {"status": "ok"}
+            finally:
+                db.close()
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        except HTTPException:
+            raise
+        except Exception as e:
+            logger.error(f"Edit message error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.post("/api/session/{session_id}/mark-stopped")
+    async def mark_stopped(request: Request, session_id: str):
+        """Mark the last assistant message as stopped by user."""
+        _verify_session_owner(request, session_id)
+        try:
+            session = session_manager.get_session(session_id)
+            # Find last assistant message and add stopped metadata
+            for msg in reversed(session.history):
+                if (isinstance(msg, ChatMessage) and msg.role == 'assistant') or \
+                   (isinstance(msg, dict) and msg.get('role') == 'assistant'):
+                    if isinstance(msg, ChatMessage):
+                        if not msg.metadata:
+                            msg.metadata = {}
+                        msg.metadata['stopped'] = True
+                        if not msg.metadata.get('model'):
+                            msg.metadata['model'] = session.model
+                    else:
+                        if 'metadata' not in msg:
+                            msg['metadata'] = {}
+                        msg['metadata']['stopped'] = True
+                        if not msg['metadata'].get('model'):
+                            msg['metadata']['model'] = session.model
+                    break
+            # Also update in DB
+            db = SessionLocal()
+            try:
+                import json as _json
+                db_messages = (
+                    db.query(DbChatMessage)
+                    .filter(DbChatMessage.session_id == session_id, DbChatMessage.role == 'assistant')
+                    .order_by(DbChatMessage.timestamp.desc())
+                    .first()
+                )
+                if db_messages:
+                    meta = {}
+                    if db_messages.meta_data:
+                        try:
+                            meta = _json.loads(db_messages.meta_data)
+                        except (json.JSONDecodeError, ValueError):
+                            pass
+                    meta['stopped'] = True
+                    if not meta.get('model'):
+                        meta['model'] = session.model
+                    db_messages.meta_data = _json.dumps(meta)
+                    db.commit()
+            finally:
+                db.close()
+            session_manager.save_sessions()
+            return {"status": "ok"}
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        except Exception as e:
+            logger.error(f"Mark stopped error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.post("/api/session/{session_id}/update-last-meta")
+    async def update_last_meta(request: Request, session_id: str):
+        """Merge metadata into the last assistant message (e.g. save variants)."""
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            meta_update = body.get("metadata", {})
+            session = session_manager.get_session(session_id)
+
+            # Update in-memory
+            for msg in reversed(session.history):
+                if (isinstance(msg, ChatMessage) and msg.role == 'assistant') or \
+                   (isinstance(msg, dict) and msg.get('role') == 'assistant'):
+                    if isinstance(msg, ChatMessage):
+                        if not msg.metadata:
+                            msg.metadata = {}
+                        msg.metadata.update(meta_update)
+                    else:
+                        if 'metadata' not in msg:
+                            msg['metadata'] = {}
+                        msg['metadata'].update(meta_update)
+                    break
+
+            # Update in DB
+            db = SessionLocal()
+            try:
+                import json as _json
+                db_msg = (
+                    db.query(DbChatMessage)
+                    .filter(DbChatMessage.session_id == session_id, DbChatMessage.role == 'assistant')
+                    .order_by(DbChatMessage.timestamp.desc())
+                    .first()
+                )
+                if db_msg:
+                    meta = {}
+                    if db_msg.meta_data:
+                        try: meta = _json.loads(db_msg.meta_data)
+                        except (json.JSONDecodeError, ValueError): pass
+                    meta.update(meta_update)
+                    db_msg.meta_data = _json.dumps(meta)
+                    db.commit()
+            finally:
+                db.close()
+            session_manager.save_sessions()
+            return {"status": "ok"}
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        except Exception as e:
+            logger.error(f"Update last meta error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.post("/api/session/{session_id}/merge-last-assistant")
+    async def merge_last_assistant(request: Request, session_id: str):
+        """Merge the last two assistant messages into one (for continue)."""
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            separator = body.get("separator", "\n\n")
+            session = session_manager.get_session(session_id)
+
+            # Find last two assistant messages in-memory
+            ai_indices = []
+            for i, msg in enumerate(session.history):
+                role = msg.role if isinstance(msg, ChatMessage) else msg.get('role', '')
+                if role == 'assistant':
+                    ai_indices.append(i)
+
+            if len(ai_indices) < 2:
+                return {"status": "ok", "merged": False}
+
+            idx1, idx2 = ai_indices[-2], ai_indices[-1]
+            msg1, msg2 = session.history[idx1], session.history[idx2]
+
+            content1 = msg1.content if isinstance(msg1, ChatMessage) else msg1.get('content', '')
+            content2 = msg2.content if isinstance(msg2, ChatMessage) else msg2.get('content', '')
+            merged_content = content1 + separator + content2
+
+            # Merge metadata
+            meta1 = (msg1.metadata if isinstance(msg1, ChatMessage) else msg1.get('metadata')) or {}
+            meta2 = (msg2.metadata if isinstance(msg2, ChatMessage) else msg2.get('metadata')) or {}
+            merged_meta = {**meta1, **meta2}
+            merged_meta.pop('stopped', None)  # no longer stopped after continue
+
+            # Update first message, remove second
+            if isinstance(msg1, ChatMessage):
+                msg1.content = merged_content
+                msg1.metadata = merged_meta
+            else:
+                msg1['content'] = merged_content
+                msg1['metadata'] = merged_meta
+
+            # Also remove the hidden "continue" user message between them if present
+            # It's the message at idx2-1 if it's a user message with continue text
+            remove_indices = [idx2]
+            if idx2 - 1 > idx1:
+                between = session.history[idx2 - 1]
+                between_role = between.role if isinstance(between, ChatMessage) else between.get('role', '')
+                between_content = between.content if isinstance(between, ChatMessage) else between.get('content', '')
+                if between_role == 'user' and 'previous response was interrupted' in between_content:
+                    remove_indices.insert(0, idx2 - 1)
+
+            for ri in sorted(remove_indices, reverse=True):
+                session.history.pop(ri)
+
+            # Update DB
+            db = SessionLocal()
+            try:
+                import json as _json
+                db_messages = (
+                    db.query(DbChatMessage)
+                    .filter(DbChatMessage.session_id == session_id)
+                    .order_by(DbChatMessage.timestamp)
+                    .all()
+                )
+                # Find last two assistant messages in DB
+                ai_db = [(i, m) for i, m in enumerate(db_messages) if m.role == 'assistant']
+                if len(ai_db) >= 2:
+                    (_, db1), (_, db2) = ai_db[-2], ai_db[-1]
+                    db1.content = merged_content
+                    db1.meta_data = _json.dumps(merged_meta)
+
+                    # Mirror the in-memory deletion: remove the second assistant
+                    # message and ONLY the "continue" user message between them
+                    # (not arbitrary tool/system/user rows). The old
+                    # range-delete destroyed every row between the two assistant
+                    # messages, desyncing the DB from the in-memory history.
+                    for _row in _merge_continue_rows_to_delete(db_messages, db1, db2):
+                        db.delete(_row)
+
+                    db.commit()
+            finally:
+                db.close()
+            session_manager.save_sessions()
+            return {"status": "ok", "merged": True}
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        except Exception as e:
+            logger.error(f"Merge assistant error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.post("/api/session/{session_id}/fork")
+    async def fork_session(request: Request, session_id: str):
+        """Create a new session with messages copied up to keep_count."""
+        _verify_session_owner(request, session_id)
+        try:
+            body = await request.json()
+            keep_count = body.get("keep_count", 0)
+
+            # Get the source session
+            source = session_manager.sessions.get(session_id)
+            if not source:
+                raise HTTPException(404, "Session not found")
+
+            # Create new session
+            new_id = str(uuid.uuid4())
+            fork_name = f"\u2ADD {source.name}"
+            new_session = session_manager.create_session(
+                session_id=new_id,
+                name=fork_name,
+                endpoint_url=source.endpoint_url,
+                model=source.model,
+                rag=False,
+                owner=getattr(source, 'owner', None),
+            )
+
+            # Copy messages up to keep_count
+            msgs_to_copy = source.history[:keep_count]
+            for msg in msgs_to_copy:
+                # Copy the metadata dict. Sharing it would let the fork's
+                # persistence (add_message -> _persist_message stamps
+                # _db_id/timestamp onto the dict) mutate the SOURCE session's
+                # in-memory messages, corrupting their _db_id and breaking
+                # edit/delete-by-id on the original conversation.
+                meta = dict(msg.metadata) if isinstance(msg.metadata, dict) else None
+                new_session.add_message(ChatMessage(msg.role, msg.content, meta))
+            try:
+                from src.event_bus import fire_event
+                fire_event("session_created", getattr(source, 'owner', None))
+            except Exception:
+                logger.debug("session_created event dispatch failed", exc_info=True)
+
+            return {
+                "status": "ok",
+                "id": new_id,
+                "name": fork_name,
+                "kept": len(msgs_to_copy),
+            }
+        except HTTPException:
+            raise
+        except Exception as e:
+            logger.error(f"Fork error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    @router.get("/api/conversations/topics")
+    async def get_conversation_topics(request: Request) -> Dict[str, Any]:
+        from src.auth_helpers import require_user
+        user = require_user(request)
+        try:
+            return analyze_topics(session_manager, owner=user or None)
+        except Exception as e:
+            raise HTTPException(500, f"Topic analysis failed: {e}")
+
+    @router.post("/api/session/{session_id}/compact")
+    async def compact_session(request: Request, session_id: str):
+        """Manually trigger context compaction for a session."""
+        _verify_session_owner(request, session_id)
+        from src.auth_helpers import effective_user
+        owner = effective_user(request)
+        try:
+            session = session_manager.get_session(session_id)
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        _reject_compact_during_active_run(session_id)
+
+        try:
+            from src.model_context import estimate_tokens, get_context_length
+            from src.llm_core import llm_call_async
+            from src.endpoint_resolver import resolve_endpoint
+
+            if len(session.history) < 6:
+                return {"status": "ok", "message": "Not enough messages to compact"}
+
+            ctx_len = get_context_length(session.endpoint_url, session.model)
+            messages_before = session.get_context_messages()
+            used_before = estimate_tokens(messages_before)
+            pct_before = round((used_before / ctx_len) * 100, 1) if ctx_len else 0
+            msg_count_before = len(session.history)
+
+            # Keep only last 4 messages, summarize the rest
+            keep_count = 4
+            older = session.history[:-keep_count]
+            recent = session.history[-keep_count:]
+
+            # Build text to summarize
+            convo_text = "\n".join(
+                f"{_message_role(m).upper()}: "
+                f"{_message_text(m)[:2000]}"
+                for m in older
+            )
+
+            # Use utility model if available
+            util_url, util_model, util_headers = resolve_endpoint("utility", owner=owner or None)
+            compact_url = util_url or session.endpoint_url
+            compact_model = util_model or session.model
+            compact_headers = util_headers if util_url else session.headers
+
+            from src.context_compactor import SELF_SUMMARY_SYSTEM_PROMPT
+            compaction_count = sum(1 for m in session.history if isinstance(m, ChatMessage) and "[Conversation summary" in (m.content or ""))
+            sys_prompt = SELF_SUMMARY_SYSTEM_PROMPT.replace("{count}", str(len(older))).replace("{n}", str(compaction_count + 1))
+            summary = await llm_call_async(
+                compact_url, compact_model,
+                [
+                    {"role": "system", "content": sys_prompt},
+                    {"role": "user", "content": convo_text},
+                ],
+                temperature=0.2, max_tokens=1024,
+                headers=compact_headers, timeout=30,
+            )
+
+            # Replace session history: summary as system message + recent messages
+            # System message holds the full summary for AI context
+            system_summary = ChatMessage(
+                role="system",
+                content=f"[Conversation summary — {len(older)} earlier messages were compacted]\n\n{summary}",
+                metadata={"compacted": True, "hidden": True},
+            )
+            # Visible assistant message just shows stats
+            summary_msg = ChatMessage(
+                role="assistant",
+                content=f"**Conversation compacted** — {len(older)} messages summarized, {len(recent)} kept.",
+                metadata={"compacted": True, "messages_removed": len(older)},
+            )
+            new_history = [system_summary, summary_msg] + list(recent)
+            session.history = new_history
+            session.message_count = len(session.history)
+            logger.info(f"Compact: session {session_id} history now has {len(session.history)} messages (was {msg_count_before})")
+
+            # Update DB: delete old messages, insert summary
+            db = SessionLocal()
+            try:
+                db_msgs = db.query(DbChatMessage).filter(
+                    DbChatMessage.session_id == session_id
+                ).order_by(DbChatMessage.timestamp).all()
+
+                # Delete all but the last keep_count
+                for m in db_msgs[:-keep_count]:
+                    db.delete(m)
+
+                # Insert system summary (hidden, for AI context) and visible summary
+                import json as _json
+                import uuid
+                from datetime import datetime, timezone
+                now = datetime.now(timezone.utc)
+                db_sys_summary = DbChatMessage(
+                    id=str(uuid.uuid4()),
+                    session_id=session_id,
+                    role="system",
+                    content=system_summary.content,
+                    meta_data=_json.dumps(system_summary.metadata),
+                    timestamp=now,
+                )
+                db.add(db_sys_summary)
+                db_summary = DbChatMessage(
+                    id=str(uuid.uuid4()),
+                    session_id=session_id,
+                    role="assistant",
+                    content=summary_msg.content,
+                    meta_data=_json.dumps(summary_msg.metadata),
+                    timestamp=now,
+                )
+                db.add(db_summary)
+
+                # Update session record
+                db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
+                if db_session:
+                    db_session.message_count = len(session.history)
+                    db_session.updated_at = datetime.now(timezone.utc)
+                db.commit()
+            finally:
+                db.close()
+
+            session_manager.save_sessions()
+
+            used_after = estimate_tokens(session.get_context_messages())
+            pct_after = round((used_after / ctx_len) * 100, 1) if ctx_len else 0
+
+            return {
+                "status": "ok",
+                "message": f"Compacted: {msg_count_before} msgs → {len(session.history)} msgs ({pct_before}% → {pct_after}%)",
+                "before": pct_before,
+                "after": pct_after,
+            }
+
+        except Exception as e:
+            logger.error(f"Manual compact error {session_id}: {e}")
+            raise HTTPException(500, str(e))
+
+    return router

routes/hwfit_routes.py

@@ -191,7 +191,7 @@ def setup_hwfit_routes():
         return detect_system(host=host, ssh_port=ssh_port, platform=platform, fresh=fresh)
 
     @router.get("/models")
-    def get_models(use_case: str = "", sort: str = "newest", limit: int = 50, search: str = "", host: str = "", quant: str = "", ctx: str = "", gpu_count: str = "", gpu_group: str = "", ssh_port: str = "", platform: str = "", fresh: bool = False, refresh_catalog: bool = False, manual_mode: str = "", manual_gpu_count: str = "", manual_vram_gb: str = "", manual_ram_gb: str = "", manual_backend: str = "", ignore_detected_gpu: bool = False, ignore_detected_ram: bool = False, fit_only: bool = False):
+    def get_models(use_case: str = "", sort: str = "newest", limit: int = 50, search: str = "", host: str = "", quant: str = "", ctx: str = "", gpu_count: str = "", gpu_group: str = "", ssh_port: str = "", platform: str = "", fresh: bool = False, manual_mode: str = "", manual_gpu_count: str = "", manual_vram_gb: str = "", manual_ram_gb: str = "", manual_backend: str = "", ignore_detected_gpu: bool = False, ignore_detected_ram: bool = False, fit_only: bool = False):
         """Rank LLM models against detected hardware and return scored results.
         gpu_count: override GPU count (0 = CPU only, 1-N = simulate N GPUs of the
             active group). gpu_group: index into system.gpu_groups (the homogeneous
@@ -200,17 +200,11 @@ def setup_hwfit_routes():
         fresh=true bypasses the hardware-detection cache."""
         from services.hwfit.hardware import detect_system
         from services.hwfit.fit import rank_models
-        from services.hwfit.models import get_models, model_catalog_path, refresh_dynamic_catalogs
+        from services.hwfit.models import get_models, model_catalog_path
         host, ssh_port = _validate_detection_target(host, ssh_port)
         system = deepcopy(detect_system(host=host, ssh_port=ssh_port, platform=platform, fresh=fresh))
         if system.get("error"):
             return {"system": system, "models": [], "error": system["error"]}
-        catalog_refresh = None
-        if refresh_catalog:
-            try:
-                catalog_refresh = refresh_dynamic_catalogs(force=True)
-            except Exception as e:
-                catalog_refresh = {"error": str(e)}
         if not get_models():
             return {
                 "system": system,
@@ -310,10 +304,7 @@ def setup_hwfit_routes():
             rank_kwargs.pop("target_context", None)
             rank_kwargs.pop("fit_only", None)
         results = rank_models(system, **rank_kwargs)
-        payload = {"system": system, "models": results}
-        if catalog_refresh is not None:
-            payload["catalog_refresh"] = catalog_refresh
-        return payload
+        return {"system": system, "models": results}
 
     @router.get("/profiles")
     def get_serve_profiles(model: str = "", model_path: str = "", host: str = "", ssh_port: str = "", platform: str = "", fresh: bool = False, serve_weights_gb: float = 0.0, serve_quant: str = ""):

routes/memory/__init__.py

@@ -1,5 +0,0 @@
-"""Memory route domain package (slice 2c, #4082/#4071).
-
-Contains memory_routes.py, migrated from the flat routes/ directory.
-Backward-compat shim at routes/memory_routes.py re-exports from here.
-"""

routes/memory/memory_routes.py

@@ -1,552 +0,0 @@
-# routes/memory_routes.py
-from fastapi import APIRouter, Form, HTTPException, Request, UploadFile, File
-from typing import Dict, Any, Optional, List
-import json
-import os
-import re
-import tempfile
-import time
-from datetime import datetime
-import logging
-
-# Leading list-marker like "1.", "12)", or "3:" plus surrounding whitespace.
-# Strips one prefix per call so import-from-LLM-output doesn't leave the
-# numbering inside the saved memory text. Bullet markers (-, *, •) are
-# also peeled here for the same reason.
-_LIST_PREFIX_RE = re.compile(r"^\s*(?:\d{1,3}[.):]\s+|[-*•]\s+)")
-
-
-def _strip_list_prefix(text: str) -> str:
-    if not text:
-        return text
-    return _LIST_PREFIX_RE.sub("", text, count=1).strip()
-
-from services.memory import MemoryManager
-from core.session_manager import SessionManager
-from src.request_models import MemoryAddRequest
-from core.database import SessionLocal
-from src.llm_core import llm_call_async
-from services.memory.memory_extractor import audit_memories
-from src.auth_helpers import get_current_user, require_user
-from src.endpoint_resolver import resolve_endpoint
-from src.task_endpoint import resolve_task_endpoint
-from src.upload_limits import read_upload_limited, MEMORY_IMPORT_MAX_BYTES
-
-logger = logging.getLogger(__name__)
-
-
-def setup_memory_routes(memory_manager: MemoryManager, session_manager: SessionManager, memory_vector=None):
-    """Set up memory-related routes."""
-    router = APIRouter(prefix="/api/memory", tags=["memory"])
-
-    def _owner(request: Request) -> Optional[str]:
-        return get_current_user(request)
-
-    def _assert_session_owner(session_obj, user):
-        """SECURITY: 404 if the caller does not own this session.
-
-        SessionManager.get_session is NOT owner-scoped — it returns any
-        session by id. These routes accept a caller-supplied session id, so
-        without this gate a user could target another tenant's session and
-        leak their chat history, their session-scoped LLM credentials, or the
-        session title. Mirrors session_routes / webhook_routes ownership.
-        """
-        if user is not None and getattr(session_obj, "owner", None) != user:
-            raise HTTPException(404, "Session not found")
-
-    def _verify_memory_owner(memory: dict, user: Optional[str]):
-        """Raise 404 if user doesn't own this memory.
-
-        SECURITY: strict ownership — previously `mem_owner and mem_owner != user`
-        allowed any user to read/edit/delete memories with an empty/null owner
-        field, which leaked legacy data across the multi-user deploy.
-        """
-        if user is None:
-            return  # Auth disabled
-        if memory.get("owner") != user:
-            raise HTTPException(404, "Memory not found")
-
-    @router.post("/debug")
-    def debug_memory_relevance(request: Request, query: str = Form(...)):
-        """Debug which memories would be triggered for a query"""
-        user = _owner(request)
-        memories = memory_manager.load(owner=user)
-        relevant = memory_manager.get_relevant_memories(query, memories, threshold=0.05)
-
-        return {
-            "query": query,
-            "total_memories": len(memories),
-            "relevant_count": len(relevant),
-            "relevant_memories": [{"text": m["text"], "category": m.get("category", "unknown")}
-                                 for m in relevant]
-        }
-
-    @router.post("/add", response_model=Dict[str, Any])
-    async def api_add_memory(
-        request: Request,
-        memory_data: Optional[MemoryAddRequest] = None
-    ):
-        """Add a new memory entry with optional category, source, and session reference."""
-        from src.auth_helpers import require_privilege
-        require_privilege(request, "can_manage_memory")
-        if memory_data is None:
-            form = await request.form()
-            memory_data = MemoryAddRequest(
-                text=form.get("text"),
-                category=form.get("category", "fact"),
-                source=form.get("source", "user"),
-                session_id=form.get("session_id")
-            )
-
-        user = _owner(request)
-        text = (memory_data.text or "").strip()
-        if not text:
-            raise HTTPException(400, "empty memory")
-        user_mem = memory_manager.load(owner=user)
-        if memory_manager.find_duplicates(text, user_mem):
-            return {"ok": True, "count": len(user_mem), "message": "Memory already exists"}
-
-        if memory_data.session_id:
-            try:
-                session_obj = session_manager.get_session(memory_data.session_id)
-            except KeyError:
-                raise HTTPException(404, "Session not found")
-            _assert_session_owner(session_obj, user)
-
-        new_entry = memory_manager.add_entry(text, memory_data.source, memory_data.category, owner=user)
-        if memory_data.session_id:
-            new_entry["session_id"] = memory_data.session_id
-        all_mem = memory_manager.load_all()
-        all_mem.append(new_entry)
-        memory_manager.save(all_mem)
-        # Sync vector index
-        if memory_vector and memory_vector.healthy:
-            memory_vector.add(new_entry["id"], text)
-        try:
-            from src.event_bus import fire_event
-            fire_event("memory_added", user)
-        except Exception:
-            logger.debug("memory_added event dispatch failed", exc_info=True)
-        return {"ok": True, "count": len([m for m in all_mem if m.get("owner") == user])}
-
-    @router.get("")
-    def api_get_memory(request: Request):
-        """Return all memory entries with their metadata."""
-        user = _owner(request)
-        return {"memory": memory_manager.load(owner=user)}
-
-    @router.post("/search")
-    def search_memories(request: Request, query: str = Form(...), session_id: str = Form(None), category: str = Form(None)):
-        """Search across all memories with optional filters."""
-        user = _owner(request)
-        memories = memory_manager.load(owner=user)
-
-        if session_id:
-            memories = [m for m in memories if m.get("session_id") == session_id]
-
-        if category:
-            memories = [m for m in memories if category in m.get("categories", [m.get("category", "")])]
-
-        relevant = memory_manager.get_relevant_memories(query, memories, threshold=0.05, max_items=20)
-
-        return {"memories": relevant, "total": len(relevant), "query": query}
-
-    @router.get("/timeline")
-    def memory_timeline(request: Request):
-        """Get memories in chronological order with source session information."""
-        user = _owner(request)
-        memories = memory_manager.load(owner=user)
-        sorted_memories = sorted(memories, key=lambda x: x.get("timestamp", 0), reverse=True)
-
-        results = []
-        for memory in sorted_memories:
-            if "timestamp" in memory:
-                try:
-                    dt = datetime.fromtimestamp(memory["timestamp"])
-                    memory["timestamp_str"] = dt.strftime("%Y-%m-%d %H:%M:%S")
-                except (ValueError, OSError, OverflowError):
-                    memory["timestamp_str"] = "Unknown"
-            else:
-                memory["timestamp_str"] = "Unknown"
-
-            session_id = memory.get("session_id")
-            if session_id and session_id in session_manager.sessions:
-                try:
-                    session = session_manager.get_session(session_id)
-                    if session:
-                        _assert_session_owner(session, user)
-                    memory["session_name"] = session.name if session else f"Session {session_id[:6]}"
-                except KeyError:
-                    memory["session_name"] = "Unknown"
-                except HTTPException as exc:
-                    if exc.status_code != 404:
-                        raise
-                    memory["session_name"] = "Unknown"
-            else:
-                memory["session_name"] = "Unknown"
-
-            results.append(memory)
-
-        return {"timeline": results, "total": len(results)}
-
-    @router.get("/by-session/{session_id}")
-    def get_memory_by_session(request: Request, session_id: str):
-        """Get all memories associated with a specific session."""
-        user = _owner(request)
-        try:
-            _session_obj = session_manager.get_session(session_id)
-        except KeyError:
-            raise HTTPException(404, f"Session {session_id} not found")
-        _assert_session_owner(_session_obj, user)
-        memories = memory_manager.load(owner=user)
-        session_memories = [m for m in memories if m.get("session_id") == session_id]
-
-        session_memories.sort(key=lambda x: x.get("timestamp", 0), reverse=True)
-
-        try:
-            session = session_manager.get_session(session_id)
-            session_name = session.name if session else f"Session {session_id[:6]}"
-        except KeyError:
-            session_name = f"Session {session_id[:6]}"
-
-        for memory in session_memories:
-            memory["session_name"] = session_name
-
-        return {
-            "session_id": session_id,
-            "session_name": session_name,
-            "memory_count": len(session_memories),
-            "memories": session_memories
-        }
-
-    @router.post("/extract")
-    async def extract_memory(request: Request, session: str = Form(...)) -> Dict[str, List[str]]:
-        """Analyze a session's chat history and return memory suggestions."""
-        require_user(request)
-        try:
-            sess = session_manager.get_session(session)
-        except KeyError:
-            raise HTTPException(404, "Session not found")
-        _assert_session_owner(sess, _owner(request))
-
-        system_msg = {
-            "role": "system",
-            "content": (
-                "You are a helpful assistant. Analyze the entire conversation history provided and extract any "
-                "useful factual statements, contacts, addresses, phone numbers, or other information that the user "
-                "might want to remember for future interactions. Return each piece of information as a JSON object "
-                "with a 'text' field. For example: [{'text': 'Alice lives at 123 Main St'}, {'text': 'Bob works at Acme Corp'}]. "
-                "Only include information that is specific and likely to be useful later."
-            ),
-        }
-        messages = [system_msg] + sess.get_context_messages()
-
-        t_url, t_model, t_headers = resolve_task_endpoint(
-            sess.endpoint_url, sess.model, sess.headers, owner=_owner(request)
-        )
-
-        try:
-            suggestion_text = await llm_call_async(
-                t_url,
-                t_model,
-                messages,
-                temperature=0.2,
-                max_tokens=500,
-                headers=t_headers,
-            )
-            try:
-                suggestions = json.loads(suggestion_text)
-                if isinstance(suggestions, list):
-                    suggestions = [s if isinstance(s, str) else s.get("text", "") for s in suggestions]
-                else:
-                    suggestions = []
-            except json.JSONDecodeError:
-                suggestions = [line.strip() for line in suggestion_text.splitlines() if line.strip()]
-
-            return {"suggestions": [s for s in suggestions if s]}
-        except Exception as e:
-            logger.error(f"LLM memory extraction failed (session {session}): {e}")
-            fallback = memory_manager.extract_memory_from_chat(sess.history, session)
-            return {"suggestions": [item["text"] for item in fallback]}
-
-    @router.post("/audit")
-    async def api_audit_memories(request: Request, session: str = Form(None)):
-        """Deduplicate and consolidate memories via LLM.
-
-        Uses task/utility/default settings through the shared resolver, with
-        the active session as fallback when no task or utility model is set.
-        Returns before and after memory counts.
-        """
-        user = _owner(request)
-        fallback_url = fallback_model = None
-        fallback_headers = None
-        if session:
-            try:
-                sess = session_manager.get_session(session)
-                _assert_session_owner(sess, user)
-                fallback_url = sess.endpoint_url
-                fallback_model = sess.model
-                fallback_headers = sess.headers
-            except KeyError:
-                pass
-
-        endpoint_url, model, headers = resolve_task_endpoint(
-            fallback_url, fallback_model, fallback_headers, owner=user
-        )
-
-        if not endpoint_url or not model:
-            raise HTTPException(400, "No default model configured — set one in Settings")
-
-        result = await audit_memories(
-            memory_manager,
-            memory_vector,
-            endpoint_url,
-            model,
-            headers,
-            owner=user,
-        )
-
-        if "error" in result and "before" not in result:
-            raise HTTPException(502, f"Audit failed: {result['error']}")
-
-        return {
-            "ok": "error" not in result,
-            "before": result.get("before", 0),
-            "after": result.get("after", 0),
-            "removed": result.get("before", 0) - result.get("after", 0),
-            # True when the audit skipped the LLM because nothing changed
-            # since the last tidy. Frontend already says "Already clean"
-            # for removed==0, so this is here for future use / debugging.
-            "already_tidy": bool(result.get("already_tidy")),
-        }
-
-    @router.post("/import")
-    async def import_memories_from_file(
-        request: Request,
-        session: str | None = Form(None),
-        file: UploadFile = File(...)
-    ):
-        """Extract memory suggestions from an uploaded file (PDF, TXT, MD, etc.)."""
-        from src.auth_helpers import require_privilege
-        require_privilege(request, "can_manage_memory")
-
-        endpoint_url = None
-        model = None
-        headers = {}
-
-        user = _owner(request)
-
-        if session:
-            try:
-                sess = session_manager.get_session(session)
-                _assert_session_owner(sess, user)
-            except KeyError:
-                sess = None
-            except HTTPException as exc:
-                if exc.status_code != 404:
-                    raise
-                sess = None
-
-            if sess is None:
-                logger.warning("Session %s not found or inaccessible, falling back to utility endpoint", session)
-                endpoint_url, model, headers = resolve_endpoint("utility", owner=user)
-            else:
-                endpoint_url, model, headers = resolve_task_endpoint(
-                    sess.endpoint_url, sess.model, sess.headers, owner=user
-                )
-        else:
-            endpoint_url, model, headers = resolve_task_endpoint(owner=user)
-    
-        if not endpoint_url or not model:
-            raise HTTPException(400, "No LLM model configured. Set a default model in Settings.")
-
-        content = await read_upload_limited(file, MEMORY_IMPORT_MAX_BYTES, "Memory import")
-        filename = file.filename or "upload"
-        _, ext = os.path.splitext(filename.lower())
-
-        allowed = {".txt", ".md", ".pdf", ".csv", ".log", ".json", ".py", ".js", ".html"}
-        if ext not in allowed:
-            raise HTTPException(400, f"Unsupported file type: {ext}")
-
-        # Extract text based on file type
-        if ext == ".pdf":
-            from src.document_processor import _process_pdf
-            with tempfile.NamedTemporaryFile(suffix=".pdf", delete=False) as tmp:
-                tmp.write(content)
-                tmp_path = tmp.name
-            try:
-                text = _process_pdf(tmp_path, owner=_owner(request))
-            finally:
-                os.unlink(tmp_path)
-        else:
-            try:
-                text = content.decode("utf-8")
-            except UnicodeDecodeError:
-                from charset_normalizer import detect
-                encoding = (detect(content) or {}).get("encoding") or "utf-8"
-                text = content.decode(encoding, errors="replace")
-
-        if not text.strip():
-            return {"suggestions": [], "message": "No readable content found"}
-
-        # Fast path: a .json upload that already looks like a memories export
-        # (list of {text, category, ...} dicts, or list of strings) round-trips
-        # directly without spending an LLM call to re-extract its own output.
-        # Without this, re-importing a memories.json from another account
-        # ran the file through the extractor, which often re-emitted the
-        # entries as a numbered list (and the numbering leaked into the
-        # `text` field).
-        if ext == ".json":
-            try:
-                parsed = json.loads(text)
-            except json.JSONDecodeError:
-                parsed = None
-            if isinstance(parsed, list) and parsed:
-                direct = []
-                for item in parsed:
-                    if isinstance(item, dict) and item.get("text"):
-                        direct.append({
-                            "text": _strip_list_prefix(str(item["text"])),
-                            "category": item.get("category") or "fact",
-                        })
-                    elif isinstance(item, str) and item.strip():
-                        direct.append({
-                            "text": _strip_list_prefix(item.strip()),
-                            "category": "fact",
-                        })
-                if direct:
-                    return {"suggestions": direct, "filename": filename}
-
-        # Truncate very long documents
-        if len(text) > 15000:
-            text = text[:15000] + "\n[Truncated]"
-
-        # Send to LLM for memory extraction
-        import_prompt = (
-            "You are a memory extraction assistant. The user uploaded a document. "
-            "Analyze the text below and extract specific, useful facts — things like "
-            "names, preferences, jobs, locations, relationships, opinions, projects, "
-            "goals, contacts, or any other personal details worth remembering.\n\n"
-            "Rules:\n"
-            "- Each fact should be a short, self-contained statement\n"
-            "- Do NOT extract generic knowledge\n"
-            "- Focus on personal, memorable information\n"
-            "- If there are no useful facts, return an empty array\n\n"
-            "Return a JSON array of objects with 'text' and 'category' fields.\n"
-            "Categories: 'identity', 'preference', 'fact', 'contact', 'project', 'goal'\n\n"
-            "Return ONLY valid JSON, no markdown fences."
-        )
-
-        try:
-            raw = await llm_call_async(
-                endpoint_url,
-                model,
-                [
-                    {"role": "system", "content": import_prompt},
-                    {"role": "user", "content": f"Document: {filename}\n\n{text}"},
-                ],
-                temperature=0.2,
-                max_tokens=2000,
-                headers=headers,
-            )
-
-            # Parse JSON
-            raw = raw.strip()
-            if raw.startswith("```"):
-                raw = raw.split("\n", 1)[-1].rsplit("```", 1)[0].strip()
-
-            suggestions = json.loads(raw)
-            if isinstance(suggestions, list):
-                normalized = []
-                for s in suggestions:
-                    if not s:
-                        continue
-                    if isinstance(s, dict):
-                        s = dict(s)
-                        if s.get("text"):
-                            s["text"] = _strip_list_prefix(str(s["text"]))
-                        normalized.append(s)
-                    else:
-                        normalized.append({"text": _strip_list_prefix(str(s)), "category": "fact"})
-                suggestions = normalized
-            else:
-                suggestions = []
-
-            return {"suggestions": suggestions, "filename": filename}
-
-        except json.JSONDecodeError:
-            # Fallback: split by lines, stripping any "1.", "2)" markdown-list
-            # numbering the model added so saved memories don't keep the prefix.
-            lines = [_strip_list_prefix(l.strip()) for l in raw.splitlines() if l.strip() and len(l.strip()) > 5]
-            return {"suggestions": [{"text": l, "category": "fact"} for l in lines[:20]], "filename": filename}
-        except Exception as e:
-            logger.error(f"Memory import extraction failed: {e}")
-            raise HTTPException(502, f"LLM extraction failed: {str(e)}")
-
-    @router.post("/{memory_id}/pin")
-    def pin_memory(request: Request, memory_id: str, pinned: bool = Form(True)):
-        """Pin or unpin a memory. Pinned memories are always included in context."""
-        user = _owner(request)
-        all_mem = memory_manager.load_all()
-        for i, memory in enumerate(all_mem):
-            if memory["id"] == memory_id:
-                _verify_memory_owner(memory, user)
-                all_mem[i]["pinned"] = pinned
-                memory_manager.save(all_mem)
-                return {"ok": True, "pinned": pinned}
-        raise HTTPException(404, f"Memory item {memory_id} not found")
-
-    # Wildcard routes MUST come last — otherwise they swallow /import, /search, etc.
-    @router.get("/{memory_id}")
-    def get_memory_item(request: Request, memory_id: str):
-        """Get a specific memory item by ID."""
-        user = _owner(request)
-        memories = memory_manager.load(owner=user)
-        for memory in memories:
-            if memory["id"] == memory_id:
-                return {"memory": memory}
-
-        raise HTTPException(404, "Memory not found")
-
-    @router.put("/{memory_id}")
-    def update_memory(request: Request, memory_id: str, text: str = Form(...), category: str = Form(None)):
-        """Update an existing memory item with new text and optional category."""
-        user = _owner(request)
-        all_mem = memory_manager.load_all()
-        for i, memory in enumerate(all_mem):
-            if memory["id"] == memory_id:
-                _verify_memory_owner(memory, user)
-                all_mem[i]["text"] = text.strip()
-                if category:
-                    all_mem[i]["category"] = category
-                all_mem[i]["timestamp"] = int(time.time())
-
-                memory_manager.save(all_mem)
-                # Sync vector index (remove old, add updated)
-                if memory_vector and memory_vector.healthy:
-                    memory_vector.remove(memory_id)
-                    memory_vector.add(memory_id, text.strip())
-                return {"ok": True, "message": "Memory updated successfully"}
-
-        raise HTTPException(404, f"Memory item {memory_id} not found")
-
-    @router.delete("/{memory_id}")
-    def delete_memory(request: Request, memory_id: str):
-        """Delete a memory item by its ID."""
-        user = _owner(request)
-        all_mem = memory_manager.load_all()
-
-        # Find and verify ownership before deleting
-        target = next((m for m in all_mem if m["id"] == memory_id), None)
-        if not target:
-            raise HTTPException(404, f"Memory item {memory_id} not found")
-        _verify_memory_owner(target, user)
-
-        all_mem = [m for m in all_mem if m["id"] != memory_id]
-        memory_manager.save(all_mem)
-        # Sync vector index
-        if memory_vector and memory_vector.healthy:
-            memory_vector.remove(memory_id)
-        return {"ok": True, "message": "Memory deleted successfully"}
-
-    return router

routes/memory_routes.py

@@ -1,18 +1,552 @@
-"""Backward-compat shim — canonical location is routes/memory/memory_routes.py.
+# routes/memory_routes.py
+from fastapi import APIRouter, Form, HTTPException, Request, UploadFile, File
+from typing import Dict, Any, Optional, List
+import json
+import os
+import re
+import tempfile
+import time
+from datetime import datetime
+import logging
 
-This module is replaced in ``sys.modules`` by the canonical module object so
-that ``import routes.memory_routes``, ``from routes.memory_routes import X``,
-``importlib.import_module("routes.memory_routes")``, and
-``monkeypatch.setattr(routes.memory_routes, "ATTR", ...)`` (used by
-test_memory_routes_session_owner.py and test_memory_owner_isolation.py via
-``import ... as mr`` + ``setattr(mr, ...)``) all operate on the *same* object
-the application actually uses. Keeps existing import paths working after
-slice 2c (#4082/#4071). Source-introspection tests read the canonical file
-by path.
-"""
+# Leading list-marker like "1.", "12)", or "3:" plus surrounding whitespace.
+# Strips one prefix per call so import-from-LLM-output doesn't leave the
+# numbering inside the saved memory text. Bullet markers (-, *, •) are
+# also peeled here for the same reason.
+_LIST_PREFIX_RE = re.compile(r"^\s*(?:\d{1,3}[.):]\s+|[-*•]\s+)")
 
-import sys as _sys
 
-from routes.memory import memory_routes as _canonical  # noqa: F401
+def _strip_list_prefix(text: str) -> str:
+    if not text:
+        return text
+    return _LIST_PREFIX_RE.sub("", text, count=1).strip()
 
-_sys.modules[__name__] = _canonical
+from services.memory import MemoryManager
+from core.session_manager import SessionManager
+from src.request_models import MemoryAddRequest
+from core.database import SessionLocal
+from src.llm_core import llm_call_async
+from services.memory.memory_extractor import audit_memories
+from src.auth_helpers import get_current_user, require_user
+from src.endpoint_resolver import resolve_endpoint
+from src.task_endpoint import resolve_task_endpoint
+from src.upload_limits import read_upload_limited, MEMORY_IMPORT_MAX_BYTES
+
+logger = logging.getLogger(__name__)
+
+
+def setup_memory_routes(memory_manager: MemoryManager, session_manager: SessionManager, memory_vector=None):
+    """Set up memory-related routes."""
+    router = APIRouter(prefix="/api/memory", tags=["memory"])
+
+    def _owner(request: Request) -> Optional[str]:
+        return get_current_user(request)
+
+    def _assert_session_owner(session_obj, user):
+        """SECURITY: 404 if the caller does not own this session.
+
+        SessionManager.get_session is NOT owner-scoped — it returns any
+        session by id. These routes accept a caller-supplied session id, so
+        without this gate a user could target another tenant's session and
+        leak their chat history, their session-scoped LLM credentials, or the
+        session title. Mirrors session_routes / webhook_routes ownership.
+        """
+        if user is not None and getattr(session_obj, "owner", None) != user:
+            raise HTTPException(404, "Session not found")
+
+    def _verify_memory_owner(memory: dict, user: Optional[str]):
+        """Raise 404 if user doesn't own this memory.
+
+        SECURITY: strict ownership — previously `mem_owner and mem_owner != user`
+        allowed any user to read/edit/delete memories with an empty/null owner
+        field, which leaked legacy data across the multi-user deploy.
+        """
+        if user is None:
+            return  # Auth disabled
+        if memory.get("owner") != user:
+            raise HTTPException(404, "Memory not found")
+
+    @router.post("/debug")
+    def debug_memory_relevance(request: Request, query: str = Form(...)):
+        """Debug which memories would be triggered for a query"""
+        user = _owner(request)
+        memories = memory_manager.load(owner=user)
+        relevant = memory_manager.get_relevant_memories(query, memories, threshold=0.05)
+
+        return {
+            "query": query,
+            "total_memories": len(memories),
+            "relevant_count": len(relevant),
+            "relevant_memories": [{"text": m["text"], "category": m.get("category", "unknown")}
+                                 for m in relevant]
+        }
+
+    @router.post("/add", response_model=Dict[str, Any])
+    async def api_add_memory(
+        request: Request,
+        memory_data: Optional[MemoryAddRequest] = None
+    ):
+        """Add a new memory entry with optional category, source, and session reference."""
+        from src.auth_helpers import require_privilege
+        require_privilege(request, "can_manage_memory")
+        if memory_data is None:
+            form = await request.form()
+            memory_data = MemoryAddRequest(
+                text=form.get("text"),
+                category=form.get("category", "fact"),
+                source=form.get("source", "user"),
+                session_id=form.get("session_id")
+            )
+
+        user = _owner(request)
+        text = (memory_data.text or "").strip()
+        if not text:
+            raise HTTPException(400, "empty memory")
+        user_mem = memory_manager.load(owner=user)
+        if memory_manager.find_duplicates(text, user_mem):
+            return {"ok": True, "count": len(user_mem), "message": "Memory already exists"}
+
+        if memory_data.session_id:
+            try:
+                session_obj = session_manager.get_session(memory_data.session_id)
+            except KeyError:
+                raise HTTPException(404, "Session not found")
+            _assert_session_owner(session_obj, user)
+
+        new_entry = memory_manager.add_entry(text, memory_data.source, memory_data.category, owner=user)
+        if memory_data.session_id:
+            new_entry["session_id"] = memory_data.session_id
+        all_mem = memory_manager.load_all()
+        all_mem.append(new_entry)
+        memory_manager.save(all_mem)
+        # Sync vector index
+        if memory_vector and memory_vector.healthy:
+            memory_vector.add(new_entry["id"], text)
+        try:
+            from src.event_bus import fire_event
+            fire_event("memory_added", user)
+        except Exception:
+            logger.debug("memory_added event dispatch failed", exc_info=True)
+        return {"ok": True, "count": len([m for m in all_mem if m.get("owner") == user])}
+
+    @router.get("")
+    def api_get_memory(request: Request):
+        """Return all memory entries with their metadata."""
+        user = _owner(request)
+        return {"memory": memory_manager.load(owner=user)}
+
+    @router.post("/search")
+    def search_memories(request: Request, query: str = Form(...), session_id: str = Form(None), category: str = Form(None)):
+        """Search across all memories with optional filters."""
+        user = _owner(request)
+        memories = memory_manager.load(owner=user)
+
+        if session_id:
+            memories = [m for m in memories if m.get("session_id") == session_id]
+
+        if category:
+            memories = [m for m in memories if category in m.get("categories", [m.get("category", "")])]
+
+        relevant = memory_manager.get_relevant_memories(query, memories, threshold=0.05, max_items=20)
+
+        return {"memories": relevant, "total": len(relevant), "query": query}
+
+    @router.get("/timeline")
+    def memory_timeline(request: Request):
+        """Get memories in chronological order with source session information."""
+        user = _owner(request)
+        memories = memory_manager.load(owner=user)
+        sorted_memories = sorted(memories, key=lambda x: x.get("timestamp", 0), reverse=True)
+
+        results = []
+        for memory in sorted_memories:
+            if "timestamp" in memory:
+                try:
+                    dt = datetime.fromtimestamp(memory["timestamp"])
+                    memory["timestamp_str"] = dt.strftime("%Y-%m-%d %H:%M:%S")
+                except (ValueError, OSError, OverflowError):
+                    memory["timestamp_str"] = "Unknown"
+            else:
+                memory["timestamp_str"] = "Unknown"
+
+            session_id = memory.get("session_id")
+            if session_id and session_id in session_manager.sessions:
+                try:
+                    session = session_manager.get_session(session_id)
+                    if session:
+                        _assert_session_owner(session, user)
+                    memory["session_name"] = session.name if session else f"Session {session_id[:6]}"
+                except KeyError:
+                    memory["session_name"] = "Unknown"
+                except HTTPException as exc:
+                    if exc.status_code != 404:
+                        raise
+                    memory["session_name"] = "Unknown"
+            else:
+                memory["session_name"] = "Unknown"
+
+            results.append(memory)
+
+        return {"timeline": results, "total": len(results)}
+
+    @router.get("/by-session/{session_id}")
+    def get_memory_by_session(request: Request, session_id: str):
+        """Get all memories associated with a specific session."""
+        user = _owner(request)
+        try:
+            _session_obj = session_manager.get_session(session_id)
+        except KeyError:
+            raise HTTPException(404, f"Session {session_id} not found")
+        _assert_session_owner(_session_obj, user)
+        memories = memory_manager.load(owner=user)
+        session_memories = [m for m in memories if m.get("session_id") == session_id]
+
+        session_memories.sort(key=lambda x: x.get("timestamp", 0), reverse=True)
+
+        try:
+            session = session_manager.get_session(session_id)
+            session_name = session.name if session else f"Session {session_id[:6]}"
+        except KeyError:
+            session_name = f"Session {session_id[:6]}"
+
+        for memory in session_memories:
+            memory["session_name"] = session_name
+
+        return {
+            "session_id": session_id,
+            "session_name": session_name,
+            "memory_count": len(session_memories),
+            "memories": session_memories
+        }
+
+    @router.post("/extract")
+    async def extract_memory(request: Request, session: str = Form(...)) -> Dict[str, List[str]]:
+        """Analyze a session's chat history and return memory suggestions."""
+        require_user(request)
+        try:
+            sess = session_manager.get_session(session)
+        except KeyError:
+            raise HTTPException(404, "Session not found")
+        _assert_session_owner(sess, _owner(request))
+
+        system_msg = {
+            "role": "system",
+            "content": (
+                "You are a helpful assistant. Analyze the entire conversation history provided and extract any "
+                "useful factual statements, contacts, addresses, phone numbers, or other information that the user "
+                "might want to remember for future interactions. Return each piece of information as a JSON object "
+                "with a 'text' field. For example: [{'text': 'Alice lives at 123 Main St'}, {'text': 'Bob works at Acme Corp'}]. "
+                "Only include information that is specific and likely to be useful later."
+            ),
+        }
+        messages = [system_msg] + sess.get_context_messages()
+
+        t_url, t_model, t_headers = resolve_task_endpoint(
+            sess.endpoint_url, sess.model, sess.headers, owner=_owner(request)
+        )
+
+        try:
+            suggestion_text = await llm_call_async(
+                t_url,
+                t_model,
+                messages,
+                temperature=0.2,
+                max_tokens=500,
+                headers=t_headers,
+            )
+            try:
+                suggestions = json.loads(suggestion_text)
+                if isinstance(suggestions, list):
+                    suggestions = [s if isinstance(s, str) else s.get("text", "") for s in suggestions]
+                else:
+                    suggestions = []
+            except json.JSONDecodeError:
+                suggestions = [line.strip() for line in suggestion_text.splitlines() if line.strip()]
+
+            return {"suggestions": [s for s in suggestions if s]}
+        except Exception as e:
+            logger.error(f"LLM memory extraction failed (session {session}): {e}")
+            fallback = memory_manager.extract_memory_from_chat(sess.history, session)
+            return {"suggestions": [item["text"] for item in fallback]}
+
+    @router.post("/audit")
+    async def api_audit_memories(request: Request, session: str = Form(None)):
+        """Deduplicate and consolidate memories via LLM.
+
+        Uses task/utility/default settings through the shared resolver, with
+        the active session as fallback when no task or utility model is set.
+        Returns before and after memory counts.
+        """
+        user = _owner(request)
+        fallback_url = fallback_model = None
+        fallback_headers = None
+        if session:
+            try:
+                sess = session_manager.get_session(session)
+                _assert_session_owner(sess, user)
+                fallback_url = sess.endpoint_url
+                fallback_model = sess.model
+                fallback_headers = sess.headers
+            except KeyError:
+                pass
+
+        endpoint_url, model, headers = resolve_task_endpoint(
+            fallback_url, fallback_model, fallback_headers, owner=user
+        )
+
+        if not endpoint_url or not model:
+            raise HTTPException(400, "No default model configured — set one in Settings")
+
+        result = await audit_memories(
+            memory_manager,
+            memory_vector,
+            endpoint_url,
+            model,
+            headers,
+            owner=user,
+        )
+
+        if "error" in result and "before" not in result:
+            raise HTTPException(502, f"Audit failed: {result['error']}")
+
+        return {
+            "ok": "error" not in result,
+            "before": result.get("before", 0),
+            "after": result.get("after", 0),
+            "removed": result.get("before", 0) - result.get("after", 0),
+            # True when the audit skipped the LLM because nothing changed
+            # since the last tidy. Frontend already says "Already clean"
+            # for removed==0, so this is here for future use / debugging.
+            "already_tidy": bool(result.get("already_tidy")),
+        }
+
+    @router.post("/import")
+    async def import_memories_from_file(
+        request: Request,
+        session: str | None = Form(None),
+        file: UploadFile = File(...)
+    ):
+        """Extract memory suggestions from an uploaded file (PDF, TXT, MD, etc.)."""
+        from src.auth_helpers import require_privilege
+        require_privilege(request, "can_manage_memory")
+
+        endpoint_url = None
+        model = None
+        headers = {}
+
+        user = _owner(request)
+
+        if session:
+            try:
+                sess = session_manager.get_session(session)
+                _assert_session_owner(sess, user)
+            except KeyError:
+                sess = None
+            except HTTPException as exc:
+                if exc.status_code != 404:
+                    raise
+                sess = None
+
+            if sess is None:
+                logger.warning("Session %s not found or inaccessible, falling back to utility endpoint", session)
+                endpoint_url, model, headers = resolve_endpoint("utility", owner=user)
+            else:
+                endpoint_url, model, headers = resolve_task_endpoint(
+                    sess.endpoint_url, sess.model, sess.headers, owner=user
+                )
+        else:
+            endpoint_url, model, headers = resolve_task_endpoint(owner=user)
+    
+        if not endpoint_url or not model:
+            raise HTTPException(400, "No LLM model configured. Set a default model in Settings.")
+
+        content = await read_upload_limited(file, MEMORY_IMPORT_MAX_BYTES, "Memory import")
+        filename = file.filename or "upload"
+        _, ext = os.path.splitext(filename.lower())
+
+        allowed = {".txt", ".md", ".pdf", ".csv", ".log", ".json", ".py", ".js", ".html"}
+        if ext not in allowed:
+            raise HTTPException(400, f"Unsupported file type: {ext}")
+
+        # Extract text based on file type
+        if ext == ".pdf":
+            from src.document_processor import _process_pdf
+            with tempfile.NamedTemporaryFile(suffix=".pdf", delete=False) as tmp:
+                tmp.write(content)
+                tmp_path = tmp.name
+            try:
+                text = _process_pdf(tmp_path, owner=_owner(request))
+            finally:
+                os.unlink(tmp_path)
+        else:
+            try:
+                text = content.decode("utf-8")
+            except UnicodeDecodeError:
+                from charset_normalizer import detect
+                encoding = (detect(content) or {}).get("encoding") or "utf-8"
+                text = content.decode(encoding, errors="replace")
+
+        if not text.strip():
+            return {"suggestions": [], "message": "No readable content found"}
+
+        # Fast path: a .json upload that already looks like a memories export
+        # (list of {text, category, ...} dicts, or list of strings) round-trips
+        # directly without spending an LLM call to re-extract its own output.
+        # Without this, re-importing a memories.json from another account
+        # ran the file through the extractor, which often re-emitted the
+        # entries as a numbered list (and the numbering leaked into the
+        # `text` field).
+        if ext == ".json":
+            try:
+                parsed = json.loads(text)
+            except json.JSONDecodeError:
+                parsed = None
+            if isinstance(parsed, list) and parsed:
+                direct = []
+                for item in parsed:
+                    if isinstance(item, dict) and item.get("text"):
+                        direct.append({
+                            "text": _strip_list_prefix(str(item["text"])),
+                            "category": item.get("category") or "fact",
+                        })
+                    elif isinstance(item, str) and item.strip():
+                        direct.append({
+                            "text": _strip_list_prefix(item.strip()),
+                            "category": "fact",
+                        })
+                if direct:
+                    return {"suggestions": direct, "filename": filename}
+
+        # Truncate very long documents
+        if len(text) > 15000:
+            text = text[:15000] + "\n[Truncated]"
+
+        # Send to LLM for memory extraction
+        import_prompt = (
+            "You are a memory extraction assistant. The user uploaded a document. "
+            "Analyze the text below and extract specific, useful facts — things like "
+            "names, preferences, jobs, locations, relationships, opinions, projects, "
+            "goals, contacts, or any other personal details worth remembering.\n\n"
+            "Rules:\n"
+            "- Each fact should be a short, self-contained statement\n"
+            "- Do NOT extract generic knowledge\n"
+            "- Focus on personal, memorable information\n"
+            "- If there are no useful facts, return an empty array\n\n"
+            "Return a JSON array of objects with 'text' and 'category' fields.\n"
+            "Categories: 'identity', 'preference', 'fact', 'contact', 'project', 'goal'\n\n"
+            "Return ONLY valid JSON, no markdown fences."
+        )
+
+        try:
+            raw = await llm_call_async(
+                endpoint_url,
+                model,
+                [
+                    {"role": "system", "content": import_prompt},
+                    {"role": "user", "content": f"Document: {filename}\n\n{text}"},
+                ],
+                temperature=0.2,
+                max_tokens=2000,
+                headers=headers,
+            )
+
+            # Parse JSON
+            raw = raw.strip()
+            if raw.startswith("```"):
+                raw = raw.split("\n", 1)[-1].rsplit("```", 1)[0].strip()
+
+            suggestions = json.loads(raw)
+            if isinstance(suggestions, list):
+                normalized = []
+                for s in suggestions:
+                    if not s:
+                        continue
+                    if isinstance(s, dict):
+                        s = dict(s)
+                        if s.get("text"):
+                            s["text"] = _strip_list_prefix(str(s["text"]))
+                        normalized.append(s)
+                    else:
+                        normalized.append({"text": _strip_list_prefix(str(s)), "category": "fact"})
+                suggestions = normalized
+            else:
+                suggestions = []
+
+            return {"suggestions": suggestions, "filename": filename}
+
+        except json.JSONDecodeError:
+            # Fallback: split by lines, stripping any "1.", "2)" markdown-list
+            # numbering the model added so saved memories don't keep the prefix.
+            lines = [_strip_list_prefix(l.strip()) for l in raw.splitlines() if l.strip() and len(l.strip()) > 5]
+            return {"suggestions": [{"text": l, "category": "fact"} for l in lines[:20]], "filename": filename}
+        except Exception as e:
+            logger.error(f"Memory import extraction failed: {e}")
+            raise HTTPException(502, f"LLM extraction failed: {str(e)}")
+
+    @router.post("/{memory_id}/pin")
+    def pin_memory(request: Request, memory_id: str, pinned: bool = Form(True)):
+        """Pin or unpin a memory. Pinned memories are always included in context."""
+        user = _owner(request)
+        all_mem = memory_manager.load_all()
+        for i, memory in enumerate(all_mem):
+            if memory["id"] == memory_id:
+                _verify_memory_owner(memory, user)
+                all_mem[i]["pinned"] = pinned
+                memory_manager.save(all_mem)
+                return {"ok": True, "pinned": pinned}
+        raise HTTPException(404, f"Memory item {memory_id} not found")
+
+    # Wildcard routes MUST come last — otherwise they swallow /import, /search, etc.
+    @router.get("/{memory_id}")
+    def get_memory_item(request: Request, memory_id: str):
+        """Get a specific memory item by ID."""
+        user = _owner(request)
+        memories = memory_manager.load(owner=user)
+        for memory in memories:
+            if memory["id"] == memory_id:
+                return {"memory": memory}
+
+        raise HTTPException(404, "Memory not found")
+
+    @router.put("/{memory_id}")
+    def update_memory(request: Request, memory_id: str, text: str = Form(...), category: str = Form(None)):
+        """Update an existing memory item with new text and optional category."""
+        user = _owner(request)
+        all_mem = memory_manager.load_all()
+        for i, memory in enumerate(all_mem):
+            if memory["id"] == memory_id:
+                _verify_memory_owner(memory, user)
+                all_mem[i]["text"] = text.strip()
+                if category:
+                    all_mem[i]["category"] = category
+                all_mem[i]["timestamp"] = int(time.time())
+
+                memory_manager.save(all_mem)
+                # Sync vector index (remove old, add updated)
+                if memory_vector and memory_vector.healthy:
+                    memory_vector.remove(memory_id)
+                    memory_vector.add(memory_id, text.strip())
+                return {"ok": True, "message": "Memory updated successfully"}
+
+        raise HTTPException(404, f"Memory item {memory_id} not found")
+
+    @router.delete("/{memory_id}")
+    def delete_memory(request: Request, memory_id: str):
+        """Delete a memory item by its ID."""
+        user = _owner(request)
+        all_mem = memory_manager.load_all()
+
+        # Find and verify ownership before deleting
+        target = next((m for m in all_mem if m["id"] == memory_id), None)
+        if not target:
+            raise HTTPException(404, f"Memory item {memory_id} not found")
+        _verify_memory_owner(target, user)
+
+        all_mem = [m for m in all_mem if m["id"] != memory_id]
+        memory_manager.save(all_mem)
+        # Sync vector index
+        if memory_vector and memory_vector.healthy:
+            memory_vector.remove(memory_id)
+        return {"ok": True, "message": "Memory deleted successfully"}
+
+    return router

routes/model_routes.py

@@ -19,7 +19,6 @@ from fastapi.responses import StreamingResponse
 from core.database import SessionLocal, ModelEndpoint, Session as DbSession
 from core.log_safety import redact_url as _redact_url_for_log
 from core.middleware import require_admin
-from src.constants import COOKBOOK_STATE_FILE
 from src.llm_core import _detect_provider, _host_match, ANTHROPIC_MODELS
 from src.tls_overrides import llm_verify
 from src.settings import load_settings as _load_settings, save_settings as _save_settings
@@ -113,67 +112,6 @@ def _clear_endpoint_settings_for_endpoint(settings: dict, ep_id: str, *, include
     return cleared
 
 
-_COOKBOOK_ACTIVE_SERVE_STATUSES = {
-    "starting", "loading", "ready", "running", "restarting",
-}
-
-
-def _active_cookbook_endpoint_ids() -> set[str]:
-    """Endpoint IDs owned by active Cookbook serve tasks.
-
-    Cookbook auto-registers endpoints with ids like ``local-*``. Those rows are
-    managed lifecycle state, not durable user configuration. If a tmux stream is
-    stopped or an old task lingers, the row must stop participating in model
-    selection and defaults.
-    """
-    try:
-        if not os.path.exists(COOKBOOK_STATE_FILE):
-            return set()
-        with open(COOKBOOK_STATE_FILE, "r", encoding="utf-8") as fh:
-            raw = fh.read()
-        state = json.loads(raw)
-    except Exception:
-        return set()
-    out: set[str] = set()
-    for task in state.get("tasks") or []:
-        if not isinstance(task, dict) or task.get("type") != "serve":
-            continue
-        if str(task.get("status") or "").lower() not in _COOKBOOK_ACTIVE_SERVE_STATUSES:
-            continue
-        ep_id = task.get("_endpointId") or task.get("endpointId") or task.get("endpoint_id")
-        if ep_id:
-            out.add(str(ep_id))
-    return out
-
-
-def _disable_stale_cookbook_local_endpoints(db) -> int:
-    """Disable enabled cookbook endpoints whose serve task is no longer active."""
-    active_ids = _active_cookbook_endpoint_ids()
-    if not active_ids:
-        return 0
-    stale = (
-        db.query(ModelEndpoint)
-        .filter(ModelEndpoint.is_enabled == True)  # noqa: E712
-        .filter(ModelEndpoint.id.like("local-%"))
-        .filter(~ModelEndpoint.id.in_(active_ids))
-        .all()
-    )
-    if not stale:
-        return 0
-    settings = _load_settings()
-    touched_settings = False
-    for ep in stale:
-        ep.is_enabled = False
-        ep.model_refresh_mode = "disabled"
-        if _clear_endpoint_settings_for_endpoint(settings, ep.id):
-            touched_settings = True
-        logger.info("Disabled stale Cookbook endpoint %s (%s @ %s)", ep.id, ep.name, ep.base_url)
-    if touched_settings:
-        _save_settings(settings)
-    db.commit()
-    return len(stale)
-
-
 def _clear_user_pref_endpoint_refs(all_prefs: dict, ep_id: str) -> int:
     """Remove endpoint references from scoped or legacy-flat user preferences."""
     if not isinstance(all_prefs, dict):
@@ -187,24 +125,7 @@ def _clear_user_pref_endpoint_refs(all_prefs: dict, ep_id: str) -> int:
     return cleared_users
 
 
-def _endpoint_visible_model_ids(ep: Any) -> List[str]:
-    """Known visible model ids for an endpoint, including pinned/manual ids."""
-    if ep is None:
-        return []
-    return _visible_models(
-        getattr(ep, "cached_models", None),
-        getattr(ep, "hidden_models", None),
-        getattr(ep, "pinned_models", None),
-    )
-
-
-def _default_endpoint_needs_assignment(
-    current_default_id: str,
-    enabled_endpoint_ids,
-    *,
-    current_default_endpoint: Any = None,
-    current_default_model: str = "",
-) -> bool:
+def _default_endpoint_needs_assignment(current_default_id: str, enabled_endpoint_ids) -> bool:
     """Whether the global default chat endpoint should be (re)assigned.
 
     True when nothing is configured yet, or the configured default no longer
@@ -216,14 +137,7 @@ def _default_endpoint_needs_assignment(
     """
     if not current_default_id:
         return True
-    if current_default_id not in enabled_endpoint_ids:
-        return True
-    if current_default_endpoint is None:
-        return False
-    if not (current_default_model or "").strip():
-        return True
-    visible = _endpoint_visible_model_ids(current_default_endpoint)
-    return bool(visible and current_default_model not in visible)
+    return current_default_id not in enabled_endpoint_ids
 
 
 # Loopback hosts a user might type for a local model server (LM Studio,
@@ -817,19 +731,12 @@ def _is_loading_model_response(resp: Any) -> bool:
 
 
 def _openai_model_ids(data: Any) -> List[str]:
-    """Extract OpenAI-style model IDs.
+    """Extract OpenAI-style model IDs (``{"data": [{"id": ...}]}``).
 
-    Accepts both standard ``{"data": [{"id": ...}]}`` responses and bare
-    ``[{"id": ...}]`` lists returned by some OpenAI-compatible providers.
-    Tolerates non-dict/non-list bodies and non-string IDs, returning only
-    non-empty string IDs.
+    Tolerates a non-dict body and non-string IDs from non-compliant upstreams,
+    returning only non-empty string IDs.
     """
-    if isinstance(data, list):
-        items = data
-    elif isinstance(data, dict):
-        items = data.get("data")
-    else:
-        items = None
+    items = data.get("data") if isinstance(data, dict) else None
     return [m["id"] for m in (items or [])
             if isinstance(m, dict) and isinstance(m.get("id"), str) and m["id"]]
 
@@ -1152,36 +1059,6 @@ def _merge_model_ids(*lists):
     return out
 
 
-def _is_mlx_deepseek_v4_repo_id(model_id: str) -> bool:
-    m = str(model_id or "").lower()
-    return "mlx-community/deepseek-v4" in m
-
-
-def _is_mlx_deepseek_v4_shim_id(model_id: str) -> bool:
-    m = str(model_id or "").lower()
-    return "/.cache/odysseus/mlx-shims/deepseek-v4" in m
-
-
-def _filter_mlx_deepseek_v4_repo_when_shimmed(model_ids):
-    """Hide the broken MLX repo id when a launch-specific shim id is available.
-
-    mlx_lm.server may advertise the original HF repo id even though generation
-    only works through Odysseus' sanitized local shim. Keep the shim as the
-    submitted model id and remove the raw repo id from the picker/default list.
-    """
-    ids = list(model_ids or [])
-    has_shim = any(_is_mlx_deepseek_v4_shim_id(m) for m in ids)
-    if not has_shim:
-        return ids
-    return [m for m in ids if not _is_mlx_deepseek_v4_repo_id(m)]
-
-
-def _model_display_name(model_id: str) -> str:
-    if _is_mlx_deepseek_v4_shim_id(model_id):
-        return str(model_id or "").rstrip("/").split("/")[-1] or "DeepSeek-V4-Flash-4bit"
-    return str(model_id or "").split("/")[-1]
-
-
 def _visible_models(cached_models, hidden_models, pinned_models=None):
     """Merge cached + pinned model IDs, then filter out hidden ones.
 
@@ -1195,7 +1072,6 @@ def _visible_models(cached_models, hidden_models, pinned_models=None):
         _normalize_model_ids(cached_models),
         _normalize_model_ids(pinned_models),
     )
-    merged = _filter_mlx_deepseek_v4_repo_when_shimmed(merged)
     if not hidden_models:
         return merged
     hidden = set(_normalize_model_ids(hidden_models))
@@ -1311,8 +1187,6 @@ def setup_model_routes(model_discovery):
                 db = SessionLocal()
                 changed = False
                 try:
-                    if _disable_stale_cookbook_local_endpoints(db):
-                        changed = True
                     endpoints = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True).all()
                     now = _time.time()
                     groups: Dict[str, Dict[str, Any]] = {}
@@ -1386,8 +1260,6 @@ def setup_model_routes(model_discovery):
 
         db = SessionLocal()
         try:
-            if _disable_stale_cookbook_local_endpoints(db):
-                _invalidate_models_cache()
             q = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True)
             if owner and not is_admin:
                 # Regular users see: their own endpoints + null-owner
@@ -1427,9 +1299,9 @@ def setup_model_routes(model_discovery):
                     "port": 0,
                     "url": chat_url,
                     "models": curated,
-                    "models_display": [_model_display_name(mid) for mid in curated],
+                    "models_display": [mid.split("/")[-1] for mid in curated],
                     "models_extra": extra,
-                    "models_extra_display": [_model_display_name(mid) for mid in extra],
+                    "models_extra_display": [mid.split("/")[-1] for mid in extra],
                     "endpoint_id": ep.id,
                     "endpoint_name": ep.name,
                     "category": category,
@@ -1457,7 +1329,7 @@ def setup_model_routes(model_discovery):
         return {"hosts": [], "items": items}
 
     @router.get("/models")
-    def api_models(request: Request, refresh: bool = False, background: bool = False):
+    def api_models(request: Request, refresh: bool = False):
         """Get available models — per-user (caller sees only their endpoints +
         legacy/shared null-owner rows). Cached per-user for 30s."""
         # Require auth; "" is the unconfigured single-user mode, treated as
@@ -1499,11 +1371,8 @@ def setup_model_routes(model_discovery):
             return cache_entry["data"]
         result = _fetch_models(owner=owner, is_admin=_is_admin)
         _models_cache[_cache_key] = {"data": result, "time": now}
-        # Kick off background refresh to update caches from live endpoints.
-        # Page boot can opt out with background=false so opening Odysseus does
-        # not start endpoint probes against slow/offline model servers.
-        if background or refresh:
-            _refresh_caches_bg(force=refresh)
+        # Kick off background refresh to update caches from live endpoints
+        _refresh_caches_bg(force=refresh)
         return result
 
     # Brief cache for local-probe results so picker-open doesn't hammer
@@ -1512,7 +1381,6 @@ def setup_model_routes(model_discovery):
     # within ~8s of the user noticing.
     _LOCAL_PROBE_TTL = 8.0
     _local_probe_cache: Dict[str, Any] = {"data": None, "time": 0.0}
-    _local_probe_inflight: Dict[str, Any] = {"task": None}
 
     @router.get("/model-endpoints/probe-local")
     async def probe_local_endpoints(request: Request):
@@ -1527,72 +1395,58 @@ def setup_model_routes(model_discovery):
                 (now - _local_probe_cache["time"]) < _LOCAL_PROBE_TTL):
             return _local_probe_cache["data"]
 
-        import asyncio as _asyncio
-        task = _local_probe_inflight.get("task")
-        if task is not None and not task.done():
-            return await task
-
-        async def _compute_local_probe() -> Dict[str, Any]:
-            db = SessionLocal()
-            try:
-                if _disable_stale_cookbook_local_endpoints(db):
-                    _invalidate_models_cache()
-                endpoints = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True).all()
-                local_eps = []
-                for ep in endpoints:
-                    base = _normalize_base(ep.base_url)
-                    kind = _effective_endpoint_kind(ep, base)
-                    if _classify_endpoint(base, kind) == "local":
-                        local_eps.append((ep.id, base, ep.api_key))
-            finally:
-                db.close()
-
-            grouped: Dict[str, Dict[str, Any]] = {}
-            for ep_id, base, api_key in local_eps:
-                key = _refresh_key(base, api_key)
-                grouped.setdefault(key, {"base": base, "api_key": api_key, "endpoint_ids": []})["endpoint_ids"].append(ep_id)
-
-            async def _probe_one(data: Dict[str, Any]) -> Dict[str, Any]:
-                t0 = _time.time()
-                try:
-                    # Bumped 1.5s → 3.5s. The previous 1.5s budget was clipping
-                    # local vLLM endpoints on Tailscale links where the model
-                    # server is still loading (Qwen3.5-122B takes 2–3 min to
-                    # warm); /v1/models can take 500–2500 ms on a busy box,
-                    # which pushed _ping_endpoint's full path-discovery sweep
-                    # past the cap and marked the row offline despite the
-                    # user actively chatting with it.
-                    ping = await _asyncio.to_thread(_ping_endpoint, data["base"], data.get("api_key"), 3.5)
-                    lat = round((_time.time() - t0) * 1000)
-                    return {
-                        "alive": bool(ping.get("reachable")),
-                        "latency_ms": lat,
-                        "status_code": ping.get("status_code"),
-                        "error": ping.get("error"),
-                    }
-                except Exception as e:
-                    return {"alive": False, "latency_ms": None, "status_code": None, "error": str(e)[:120]}
-
-            results_list = await _asyncio.gather(
-                *[_probe_one(data) for data in grouped.values()],
-                return_exceptions=False,
-            )
-            results: Dict[str, Any] = {}
-            for data, r in zip(grouped.values(), results_list):
-                for eid in data["endpoint_ids"]:
-                    results[eid] = r
-
-            _local_probe_cache["data"] = results
-            _local_probe_cache["time"] = _time.time()
-            return results
-
-        task = _asyncio.create_task(_compute_local_probe())
-        _local_probe_inflight["task"] = task
+        db = SessionLocal()
         try:
-            return await task
+            endpoints = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True).all()
+            local_eps = []
+            for ep in endpoints:
+                base = _normalize_base(ep.base_url)
+                kind = _effective_endpoint_kind(ep, base)
+                if _classify_endpoint(base, kind) == "local":
+                    local_eps.append((ep.id, base, ep.api_key))
         finally:
-            if _local_probe_inflight.get("task") is task:
-                _local_probe_inflight["task"] = None
+            db.close()
+
+        grouped: Dict[str, Dict[str, Any]] = {}
+        for ep_id, base, api_key in local_eps:
+            key = _refresh_key(base, api_key)
+            grouped.setdefault(key, {"base": base, "api_key": api_key, "endpoint_ids": []})["endpoint_ids"].append(ep_id)
+
+        async def _probe_one(data: Dict[str, Any]) -> Dict[str, Any]:
+            t0 = _time.time()
+            try:
+                import asyncio as _asyncio
+                # Bumped 1.5s → 3.5s. The previous 1.5s budget was clipping
+                # local vLLM endpoints on Tailscale links where the model
+                # server is still loading (Qwen3.5-122B takes 2–3 min to
+                # warm); /v1/models can take 500–2500 ms on a busy box,
+                # which pushed _ping_endpoint's full path-discovery sweep
+                # past the cap and marked the row offline despite the
+                # user actively chatting with it.
+                ping = await _asyncio.to_thread(_ping_endpoint, data["base"], data.get("api_key"), 3.5)
+                lat = round((_time.time() - t0) * 1000)
+                return {
+                    "alive": bool(ping.get("reachable")),
+                    "latency_ms": lat,
+                    "status_code": ping.get("status_code"),
+                    "error": ping.get("error"),
+                }
+            except Exception as e:
+                return {"alive": False, "latency_ms": None, "status_code": None, "error": str(e)[:120]}
+
+        import asyncio as _asyncio
+        results_list = await _asyncio.gather(
+            *[_probe_one(data) for data in grouped.values()],
+            return_exceptions=False,
+        )
+        results: Dict[str, Any] = {}
+        for data, r in zip(grouped.values(), results_list):
+            for eid in data["endpoint_ids"]:
+                results[eid] = r
+
+        _local_probe_cache["data"] = results
+        _local_probe_cache["time"] = now
+        return results
 
     @router.get("/ping")
     def ping_endpoints(request: Request):
@@ -1775,8 +1629,6 @@ def setup_model_routes(model_discovery):
         require_admin(request)
         db = SessionLocal()
         try:
-            if _disable_stale_cookbook_local_endpoints(db):
-                _invalidate_models_cache()
             rows = db.query(ModelEndpoint).order_by(ModelEndpoint.created_at).all()
             results = []
             for r in rows:
@@ -1784,11 +1636,67 @@ def setup_model_routes(model_discovery):
                 hidden = _hidden_model_ids(r)
                 pinned = _normalize_model_ids(getattr(r, "pinned_models", None))
                 visible = _visible_models(all_models, r.hidden_models, pinned)
-                # Keep the list route cache-only. It feeds Settings →
-                # Added Models and must render immediately; explicit
-                # Refresh/Probe endpoints do the network work.
-                status = "online" if (all_models or pinned) else ("empty" if r.is_enabled else "offline")
+                # Endpoint counts as reachable if it has any model — including
+                # admin-pinned IDs that a probe would never surface.
+                status = "online" if (all_models or pinned) else "offline"
                 ping = None
+                # When cached_models is empty, do a quick reachability probe.
+                # Bumped 1.0s → 3.5s because the user reported endpoints they
+                # were ACTIVELY chatting with showed "offline" — the previous
+                # 1s timeout was clipping live cloud endpoints (DeepSeek can
+                # take 1.5–2.5s on /v1/models when their region is under load,
+                # vLLM on a remote GPU box behind SSH can also push past 1s).
+                # 3.5s still keeps the picker render snappy in the common
+                # "everything's already cached" path because this branch only
+                # runs for endpoints with an empty cached_models.
+                if not all_models and not pinned and r.is_enabled:
+                    base_for_ping = _normalize_base(r.base_url)
+                    kind_for_ping = _effective_endpoint_kind(r, base_for_ping)
+                    ping_timeout = 10.0 if _classify_endpoint(base_for_ping, kind_for_ping) == "local" else 3.5
+                    ping = _ping_endpoint(r.base_url, r.api_key, timeout=ping_timeout)
+                    if ping.get("reachable"):
+                        status = "loading" if ping.get("loading") else "empty"
+                        if ping.get("loading"):
+                            base = _normalize_base(r.base_url)
+                            kind = _effective_endpoint_kind(r, base)
+                            results.append({
+                                "id": r.id,
+                                "name": r.name,
+                                "base_url": r.base_url,
+                                "has_key": bool(r.api_key),
+                                "api_key_fingerprint": _api_key_fingerprint(r.api_key),
+                                "is_enabled": r.is_enabled,
+                                "models": visible,
+                                "pinned_models": pinned,
+                                "hidden_count": len(hidden),
+                                "online": True,
+                                "status": status,
+                                "ping_error": (ping or {}).get("error") if ping else None,
+                                "model_type": getattr(r, "model_type", None) or "llm",
+                                "supports_tools": getattr(r, "supports_tools", None),
+                                "endpoint_kind": kind,
+                                "category": _classify_endpoint(base, kind),
+                                "model_refresh_mode": _endpoint_refresh_mode(r, kind),
+                                "model_refresh_interval": getattr(r, "model_refresh_interval", None),
+                                "model_refresh_timeout": getattr(r, "model_refresh_timeout", None),
+                            })
+                            continue
+                        # Best-effort: if the probe came back reachable, try
+                        # to populate cached_models in the background so the
+                        # NEXT picker load shows "online" instead of "empty".
+                        # Failure here is silent — we already returned the
+                        # "empty" status, and the existing background refresh
+                        # path will eventually fill it in too.
+                        try:
+                            probed = _probe_endpoint(r.base_url, r.api_key, timeout=max(5, int(ping_timeout)))
+                            if probed:
+                                r.cached_models = json.dumps(probed)
+                                db.commit()
+                                all_models = probed
+                                visible = _visible_models(all_models, r.hidden_models, pinned)
+                                status = "online"
+                        except Exception as _refill_err:
+                            logger.debug(f"opportunistic cached_models refill failed for {r.id}: {_refill_err!r}")
                 base = _normalize_base(r.base_url)
                 kind = _effective_endpoint_kind(r, base)
                 results.append({
@@ -1918,14 +1826,7 @@ def setup_model_routes(model_discovery):
                 if api_key.strip() and not existing.api_key:
                     existing.api_key = api_key.strip()
                     changed = True
-                # Keep duplicate endpoint registration cheap. This path is hit
-                # by Cookbook/browser auto-register flows and can run while the
-                # user is sending a chat message. Probing a stale LAN endpoint
-                # here used to hold the request open for tens of seconds and
-                # contend with session creation, making "send" feel blocked.
-                # Explicit "require models" calls still probe; normal refresh
-                # belongs to /model-endpoints/{id}/models or /probe.
-                if require_model_list:
+                if should_probe:
                     probed_models = _probe_endpoint(
                         base_url,
                         (api_key.strip() or existing.api_key or None),
@@ -2012,18 +1913,7 @@ def setup_model_routes(model_discovery):
                     ModelEndpoint.is_enabled == True  # noqa: E712
                 ).all()
             }
-            current_default_id = settings.get("default_endpoint_id") or ""
-            current_default_ep = None
-            if current_default_id:
-                current_default_ep = db.query(ModelEndpoint).filter(
-                    ModelEndpoint.id == current_default_id
-                ).first()
-            if _default_endpoint_needs_assignment(
-                current_default_id,
-                enabled_ids,
-                current_default_endpoint=current_default_ep,
-                current_default_model=settings.get("default_model") or "",
-            ):
+            if _default_endpoint_needs_assignment(settings.get("default_endpoint_id") or "", enabled_ids):
                 from src.endpoint_resolver import _first_chat_model
                 settings["default_endpoint_id"] = ep.id
                 settings["default_model"] = _first_chat_model(model_ids) or ""

routes/note_routes.py

@@ -483,22 +483,11 @@ async def dispatch_reminder(
                 api_key = intg.get("api_key", "")
                 if api_key:
                     hdrs["Authorization"] = f"Bearer {api_key}"
-                # SSRF guard — same check (and env knob) as the webhook branch
-                # above: link-local / metadata addresses are always rejected;
-                # REMINDER_WEBHOOK_BLOCK_PRIVATE_IPS=true also blocks RFC-1918
-                # so a ntfy base_url can't be pointed at internal services.
-                import os as _os
-                from src.url_safety import check_outbound_url as _chk
-                _block = _os.getenv("REMINDER_WEBHOOK_BLOCK_PRIVATE_IPS", "false").lower() == "true"
-                _ok, _reason = _chk(f"{base}/{topic}", block_private=_block)
-                if not _ok:
-                    ntfy_error = f"ntfy URL rejected: {_reason}"
-                else:
-                    async with httpx.AsyncClient(timeout=10.0) as client:
-                        resp = await client.post(f"{base}/{topic}", content=ntfy_body, headers=hdrs)
-                        ntfy_sent = resp.is_success
-                        if not ntfy_sent:
-                            ntfy_error = f"ntfy returned HTTP {resp.status_code}"
+                async with httpx.AsyncClient(timeout=10.0) as client:
+                    resp = await client.post(f"{base}/{topic}", content=ntfy_body, headers=hdrs)
+                    ntfy_sent = resp.is_success
+                    if not ntfy_sent:
+                        ntfy_error = f"ntfy returned HTTP {resp.status_code}"
             else:
                 ntfy_error = "No enabled ntfy integration"
         except Exception as e:

routes/preset_routes.py

@@ -1,6 +1,5 @@
 """Preset routes — /api/presets GET, /api/presets/custom POST, user templates CRUD."""
 
-import asyncio
 import logging
 import uuid
 from typing import Dict, Any, List
@@ -103,7 +102,7 @@ def setup_preset_routes(preset_manager) -> APIRouter:
         try:
             model_spec = data.get("model") or ""
             user = effective_user(request)
-            url, model, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=user)
+            url, model, headers = _resolve_model(model_spec, owner=user)
             result = await llm_call_async(url, model, messages, temperature=0.8, max_tokens=500, headers=headers)
             return {"success": True, "prompt": result.strip()}
         except Exception as e:

routes/research/__init__.py

@@ -1,5 +0,0 @@
-"""Research route domain package (slice 2b, #4082/#4071).
-
-Contains research_routes.py, migrated from the flat routes/ directory.
-Backward-compat shim at routes/research_routes.py re-exports from here.
-"""

routes/research/research_routes.py

@@ -1,783 +0,0 @@
-"""Research background task routes — /api/research/*."""
-
-import asyncio
-import json
-import logging
-import re
-import uuid
-from datetime import datetime
-from pathlib import Path
-from typing import Optional
-
-from fastapi import APIRouter, HTTPException, Query, Request
-from fastapi.responses import HTMLResponse, StreamingResponse
-from pydantic import BaseModel, Field
-from core.middleware import INTERNAL_TOOL_USER
-from src.endpoint_resolver import resolve_endpoint
-from src.auth_helpers import _auth_disabled, get_current_user
-from core.auth import RESERVED_USERNAMES
-from src.constants import DEEP_RESEARCH_DIR
-
-_SESSION_ID_RE = re.compile(r"^[a-zA-Z0-9-]{1,128}$")
-
-
-def _validate_session_id(session_id: str) -> str:
-    if not _SESSION_ID_RE.fullmatch(session_id):
-        raise HTTPException(400, "Invalid session ID format")
-    return session_id
-
-
-def _research_storage_root() -> Path:
-    return Path(DEEP_RESEARCH_DIR).resolve()
-
-
-def _find_research_path(session_id: str) -> Path | None:
-    """Find a persisted research file without deriving its path from input."""
-    expected_name = f"{_validate_session_id(session_id)}.json"
-    root = _research_storage_root()
-    for stored_path in root.glob("*.json"):
-        if stored_path.name != expected_name:
-            continue
-        resolved = stored_path.resolve()
-        try:
-            resolved.relative_to(root)
-        except ValueError:
-            return None
-        if not resolved.is_file():
-            return None
-        return resolved
-    return None
-
-
-def _require_research_path(session_id: str) -> Path:
-    path = _find_research_path(session_id)
-    if path is None:
-        raise HTTPException(404, "Research not found")
-    return path
-
-
-def _find_owned_research_path(session_id: str, user: str) -> Path | None:
-    path = _find_research_path(session_id)
-    if path is None:
-        return None
-    try:
-        owner = json.loads(path.read_text(encoding="utf-8")).get("owner")
-    except Exception:
-        return None
-    if owner != user:
-        return None
-    return path
-
-
-logger = logging.getLogger(__name__)
-
-# Model-name substrings that are NOT chat/generation models — research must
-# never pick these as its model. An OpenAI-style endpoint often lists
-# `text-embedding-ada-002` etc. first in its model list, which is why research
-# was failing with "Cannot reach model 'text-embedding-ada-002'".
-_NON_CHAT_MODEL = (
-    "text-embedding", "embedding", "tts-", "whisper", "dall-e",
-    "moderation", "rerank", "reranker", "clip", "stable-diffusion",
-)
-
-_RESEARCH_IMAGE_BLOCKLIST = {
-    "cdn.shopify.com/s/files/1/0179/4388/7926/files/icon.png",
-}
-
-
-def _is_research_icon_or_logo_url(url: str) -> bool:
-    path = url.lower().split("?")[0]
-    return any(token in path for token in (
-        "/logo", "logo_", "-logo", "favicon", "apple-touch-icon",
-        "sprite", "icon-", "_icon", "/icons/", "badge",
-    ))
-
-
-def _research_thumbnail(data: dict) -> str:
-    """Pick the same first visible image the visual report uses as hero."""
-    hidden = set(data.get("hidden_images") or [])
-    seen = set()
-
-    def usable(image: str) -> bool:
-        image = str(image or "").strip()
-        if not image or image in seen or image in hidden:
-            return False
-        if not image.startswith("https://"):
-            return False
-        if image.endswith((".svg", ".ico", ".gif")):
-            return False
-        if any(blocked in image for blocked in _RESEARCH_IMAGE_BLOCKLIST):
-            return False
-        if _is_research_icon_or_logo_url(image):
-            return False
-        return True
-
-    for source in data.get("sources") or []:
-        if not isinstance(source, dict):
-            continue
-        image = str(source.get("image") or source.get("og_image") or "").strip()
-        if usable(image):
-            seen.add(image)
-            return image
-    for finding in data.get("raw_findings") or data.get("findings") or []:
-        if not isinstance(finding, dict):
-            continue
-        image = str(finding.get("image") or finding.get("og_image") or "").strip()
-        if usable(image):
-            seen.add(image)
-            return image
-    return ""
-
-
-def _first_chat_model(models) -> str:
-    """First model that isn't an embedding/tts/etc. — falls back to models[0]."""
-    for m in (models or []):
-        if not any(p in str(m).lower() for p in _NON_CHAT_MODEL):
-            return m
-    return (models[0] if models else "")
-
-
-def _resolve_research_endpoint(sess, owner: Optional[str] = None) -> tuple:
-    """Return (endpoint_url, model, headers) for Deep Research, checking admin overrides."""
-    owner = owner or getattr(sess, "owner", None) or None
-    url, model, headers = resolve_endpoint(
-        "research",
-        fallback_url=sess.endpoint_url,
-        fallback_model=sess.model,
-        fallback_headers=sess.headers,
-        owner=owner,
-    )
-    return url, model, headers
-
-
-def _owned_enabled_endpoint(db, owner, endpoint_id=None):
-    """An enabled ModelEndpoint VISIBLE to `owner` (their own rows + legacy
-    null-owner "shared" rows), optionally narrowed to a specific endpoint_id;
-    None if nothing visible matches.
-
-    Owner-scoped on purpose. ModelEndpoint is per-user (core/database.py: non-null
-    owner = private, "the model picker only shows the endpoint to that user") and
-    holds a decrypted `api_key`. /api/research/start feeds the resolved row's
-    api_key + base_url into research_handler.start_research(llm_endpoint=,
-    llm_headers=), so an UNSCOPED lookup — by the caller-supplied endpoint_id, or
-    via the bare first-enabled fallback — would let a research-privileged user
-    spend ANOTHER user's API key/quota and reach whatever internal base_url they
-    configured. Mirrors webhook_routes._first_enabled_endpoint and
-    session_routes._owned_endpoint. A null/empty owner is a no-op (single-user /
-    legacy mode).
-    """
-    from src.database import ModelEndpoint
-    from src.auth_helpers import owner_filter
-    q = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True)  # noqa: E712
-    if endpoint_id:
-        q = q.filter(ModelEndpoint.id == endpoint_id)
-    return owner_filter(q, ModelEndpoint, owner).first()
-
-
-def _resolve_endpoint_runtime(ep, owner=None, model: Optional[str] = None):
-    """Resolve a ModelEndpoint row into (chat_url, model, headers).
-
-    Mirrors endpoint_resolver.resolve_endpoint's provider-auth handling for
-    panel-selected research endpoints. ChatGPT Subscription endpoints keep
-    OAuth tokens in ProviderAuthSession, so ep.api_key is intentionally empty.
-    """
-    from src.endpoint_resolver import (
-        build_chat_url,
-        build_headers,
-        resolve_endpoint_runtime as resolve_model_endpoint_runtime,
-    )
-
-    try:
-        base, api_key = resolve_model_endpoint_runtime(ep, owner=owner)
-    except Exception as e:
-        logger.warning("Could not resolve endpoint credentials for research: %s", e)
-        return None
-
-    ep_model = (model or "").strip()
-    if not ep_model:
-        try:
-            models = json.loads(ep.cached_models) if ep.cached_models else []
-            if models:
-                ep_model = _first_chat_model(models)
-        except Exception:
-            pass
-    if not ep_model:
-        return None
-    return build_chat_url(base), ep_model, build_headers(api_key, base)
-
-
-def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
-    router = APIRouter(tags=["research"])
-
-    def _require_user(request: Request) -> str:
-        """All research endpoints require an authenticated user. Research
-        data isn't owner-scoped in the on-disk JSON yet, so we at least
-        block anonymous access. Multi-tenant deploys should additionally
-        verify the session belongs to this user."""
-        user = get_current_user(request)
-        if not user:
-            if _auth_disabled():
-                return ""
-            raise HTTPException(401, "Not authenticated")
-        return user
-
-    def _owns_in_memory(session_id: str, user: str) -> bool:
-        """Ownership check for an in-flight (in-memory) research task.
-        Falls back to the on-disk JSON if the task has already finished."""
-        entry = research_handler._active_tasks.get(session_id)
-        if entry is not None:
-            return entry.get("owner", "") == user
-        # Task no longer in memory — check the persisted JSON.
-        try:
-            return _find_owned_research_path(session_id, user) is not None
-        except HTTPException:
-            return False
-
-    def _require_owned_or_active_research_path(session_id: str, user: str) -> Path | None:
-        """Validate ownership once and return the completed on-disk path.
-
-        Active running research has no completed disk path yet. Completed
-        tasks can remain in _active_tasks after persistence, so prefer their
-        owned disk path when available. Completed disk lookups still reuse the
-        path after the ownership gate.
-        """
-        entry = research_handler._active_tasks.get(session_id)
-        if entry is not None:
-            if entry.get("owner", "") != user:
-                raise HTTPException(404, "No research found for this session")
-            if entry.get("status") != "running":
-                path = _find_owned_research_path(session_id, user)
-                if path is not None:
-                    return path
-            return None
-
-        path = _find_owned_research_path(session_id, user)
-        if path is None:
-            raise HTTPException(404, "No research found for this session")
-        return path
-
-    @router.get("/api/research/active")
-    async def research_active(request: Request):
-        """List all currently active (running) research tasks."""
-        user = _require_user(request)
-        active = []
-        for sid, entry in research_handler._active_tasks.items():
-            # SECURITY: only show this user's running tasks.
-            if entry.get("owner", "") != user:
-                continue
-            if entry.get("status") == "running":
-                active.append({
-                    "session_id": sid,
-                    "query": entry.get("query", ""),
-                    "status": "running",
-                    "progress": entry.get("progress", {}),
-                    "started_at": entry.get("started_at", 0),
-                })
-        return {"active": active}
-
-    @router.get("/api/research/status/{session_id}")
-    async def research_status(session_id: str, request: Request):
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        if not _owns_in_memory(session_id, user):
-            raise HTTPException(404, "No research found for this session")
-        status = research_handler.get_status(session_id)
-        if status is None:
-            raise HTTPException(404, "No research found for this session")
-        return status
-
-    @router.post("/api/research/cancel/{session_id}")
-    async def research_cancel(session_id: str, request: Request):
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        if not _owns_in_memory(session_id, user):
-            raise HTTPException(404, "No research found for this session")
-        cancelled = research_handler.cancel_research(session_id)
-        return {"cancelled": cancelled}
-
-    @router.post("/api/research/result/{session_id}")
-    async def research_result(session_id: str, request: Request):
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        if not _owns_in_memory(session_id, user):
-            raise HTTPException(404, "No research result available")
-        result = research_handler.get_result(session_id)
-        if result is None:
-            raise HTTPException(404, "No research result available")
-        sources = research_handler.get_sources(session_id) or []
-        raw_findings = research_handler.get_raw_findings(session_id) or []
-        research_handler.clear_result(session_id)
-        return {"result": result, "sources": sources, "raw_findings": raw_findings}
-
-    def _assert_owns_research(session_id: str, user: str) -> None:
-        """404-not-403 ownership gate for a research session's on-disk JSON.
-        Use BEFORE returning any data or mutating the file."""
-        path = _require_research_path(session_id)
-        try:
-            owner = json.loads(path.read_text(encoding="utf-8")).get("owner")
-        except Exception:
-            raise HTTPException(404, "Research not found")
-        if owner != user:
-            raise HTTPException(404, "Research not found")
-
-    @router.get("/api/research/report/{session_id}")
-    async def research_report(session_id: str, request: Request):
-        """Serve the visual HTML report for a completed research session."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        _assert_owns_research(session_id, user)
-        logger.info(f"Visual report requested for session {session_id}")
-        try:
-            html_content = research_handler.get_report_html(session_id)
-        except Exception as e:
-            logger.error(f"Visual report generation error: {e}", exc_info=True)
-            raise HTTPException(500, f"Report generation failed: {e}")
-        if html_content is None:
-            logger.warning(f"No report data found for session {session_id}")
-            raise HTTPException(404, "No visual report available for this session")
-        return HTMLResponse(content=html_content)
-
-    class HideImageRequest(BaseModel):
-        url: str
-
-    @router.post("/api/research/{session_id}/hide-image")
-    async def research_hide_image(session_id: str, body: HideImageRequest, request: Request):
-        """Mark an image URL as hidden for this research's visual report.
-        Persisted to the research JSON so subsequent /report renders skip it."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        _assert_owns_research(session_id, user)
-        ok = research_handler.hide_image(session_id, body.url)
-        if not ok:
-            raise HTTPException(404, "Research not found")
-        return {"ok": True}
-
-    @router.post("/api/research/{session_id}/unhide-images")
-    async def research_unhide_images(session_id: str, request: Request):
-        """Clear the hidden-images list for a research session."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        _assert_owns_research(session_id, user)
-        ok = research_handler.unhide_all_images(session_id)
-        if not ok:
-            raise HTTPException(404, "Research not found")
-        return {"ok": True}
-
-    @router.get("/api/research/library")
-    async def research_library(
-        request: Request,
-        search: Optional[str] = Query(None),
-        sort: str = Query("recent"),
-        limit: int = Query(50),
-        archived: bool = Query(False),
-    ):
-        user = _require_user(request)
-        """List all completed research for the Library panel."""
-        data_dir = Path(DEEP_RESEARCH_DIR)
-        items = []
-        for p in data_dir.glob("*.json"):
-            try:
-                d = json.loads(p.read_text(encoding="utf-8"))
-                # SECURITY: only show research belonging to this user. Legacy
-                # JSONs without an `owner` field are hidden — auth was the only
-                # gate before, so every user saw every other user's reports.
-                if d.get("owner") != user:
-                    continue
-                # Archived view shows ONLY archived reports; default hides them.
-                if bool(d.get("archived")) != archived:
-                    continue
-                query = d.get("query", "")
-                if search and search.lower() not in query.lower():
-                    continue
-                sources = d.get("sources", [])
-                items.append({
-                    "id": p.stem,
-                    "query": query,
-                    "category": d.get("category") or "",
-                    "source_count": len(sources),
-                    "status": d.get("status", "done"),
-                    "duration": d.get("stats", {}).get("Duration", ""),
-                    "rounds": d.get("stats", {}).get("Rounds", ""),
-                    "started_at": d.get("started_at", 0),
-                    "completed_at": d.get("completed_at", 0),
-                    "archived": bool(d.get("archived")),
-                    "thumbnail": _research_thumbnail(d),
-                })
-            except Exception:
-                continue
-
-        # Sort
-        if sort == "recent":
-            items.sort(key=lambda x: x["completed_at"] or 0, reverse=True)
-        elif sort == "oldest":
-            items.sort(key=lambda x: x["completed_at"] or 0)
-        elif sort == "most-messages":
-            items.sort(key=lambda x: x["source_count"], reverse=True)
-        elif sort == "alpha":
-            items.sort(key=lambda x: x["query"].lower())
-
-        return {"research": items[:limit], "total": len(items)}
-
-    @router.get("/api/research/detail/{session_id}")
-    async def research_detail(session_id: str, request: Request):
-        """Return the full JSON for a single research result — sources,
-        summary, stats — used by the Library preview panel."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        path = _require_research_path(session_id)
-        try:
-            data = json.loads(path.read_text(encoding="utf-8"))
-        except Exception as e:
-            raise HTTPException(500, f"Failed to read research: {e}")
-        # SECURITY: 404 (not 403) so we don't leak that the report exists.
-        if data.get("owner") != user:
-            raise HTTPException(404, "Research not found")
-        return data
-
-    @router.post("/api/research/{session_id}/archive")
-    async def research_archive(session_id: str, request: Request, archived: bool = Query(True)):
-        """Soft-archive / restore a research report (sets `archived` in its JSON)."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        path = _require_research_path(session_id)
-        try:
-            data = json.loads(path.read_text(encoding="utf-8"))
-            if data.get("owner") != user:
-                raise HTTPException(404, "Research not found")
-            data["archived"] = bool(archived)
-            path.write_text(json.dumps(data), encoding="utf-8")
-        except HTTPException:
-            raise
-        except Exception as e:
-            raise HTTPException(500, f"Failed to update research: {e}")
-        return {"ok": True, "id": session_id, "archived": bool(archived)}
-
-    @router.delete("/api/research/{session_id}")
-    async def research_delete(session_id: str, request: Request):
-        """Delete a research result from disk."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        json_path = _find_research_path(session_id)
-        deleted = False
-        if json_path is not None:
-            # SECURITY: verify ownership before letting the caller delete it.
-            try:
-                data = json.loads(json_path.read_text(encoding="utf-8"))
-                if data.get("owner") != user:
-                    raise HTTPException(404, "Research not found")
-            except HTTPException:
-                raise
-            except Exception:
-                raise HTTPException(404, "Research not found")
-            json_path.unlink()
-            deleted = True
-        return {"deleted": deleted}
-
-    # ------------------------------------------------------------------
-    # Panel endpoints — launch research without a chat session
-    # ------------------------------------------------------------------
-
-    class ResearchStartRequest(BaseModel):
-        query: str
-        # max_rounds=0 means "Auto" — let the AI decide when to stop, capped at 20.
-        max_rounds: int = Field(default=0, ge=0, le=20)
-        search_provider: Optional[str] = None
-        endpoint_id: Optional[str] = None
-        model: Optional[str] = None
-        max_time: int = Field(default=300, ge=60, le=1800)
-        extraction_timeout: Optional[int] = Field(default=None, ge=15, le=3600)
-        extraction_concurrency: Optional[int] = Field(default=None, ge=1, le=12)
-        category: Optional[str] = None
-
-    @router.post("/api/research/start")
-    async def research_start(body: ResearchStartRequest, request: Request):
-        """Launch a research job from the dedicated panel."""
-        from src.auth_helpers import require_privilege
-        user = require_privilege(request, "can_use_research")
-        if user == INTERNAL_TOOL_USER:
-            tool_owner = (request.headers.get("X-Odysseus-Owner") or "").strip()
-            if tool_owner and tool_owner not in RESERVED_USERNAMES:
-                auth_mgr = getattr(request.app.state, "auth_manager", None)
-                if auth_mgr is not None and getattr(auth_mgr, "is_configured", False):
-                    try:
-                        privs = auth_mgr.get_privileges(tool_owner) or {}
-                        if not privs.get("can_use_research", True):
-                            raise HTTPException(403, f"Your account is not allowed to can use research.")
-                    except HTTPException:
-                        raise
-                    except Exception:
-                        pass
-                user = tool_owner
-        session_id = f"rp-{uuid.uuid4().hex[:12]}"
-
-        if body.endpoint_id:
-            from src.database import SessionLocal
-            db = SessionLocal()
-            try:
-                # Owner-scoped: never resolve another user's private endpoint
-                # (and its decrypted api_key / internal base_url). A scoped miss
-                # reads as 404 so the endpoint's existence isn't revealed.
-                ep = _owned_enabled_endpoint(db, user, body.endpoint_id)
-                if not ep:
-                    raise HTTPException(404, "Endpoint not found or disabled")
-                resolved = _resolve_endpoint_runtime(ep, owner=user, model=body.model)
-                if not resolved:
-                    raise HTTPException(400, "Endpoint is not configured with a usable model.")
-                ep_url, ep_model, ep_headers = resolved
-            finally:
-                db.close()
-        else:
-            ep_url, ep_model, ep_headers = resolve_endpoint("research", owner=user)
-            if not ep_url:
-                ep_url, ep_model, ep_headers = resolve_endpoint("utility", owner=user)
-            # When neither research nor utility is configured, use the user's
-            # configured DEFAULT model (default_endpoint_id/default_model) rather
-            # than arbitrarily grabbing the first enabled endpoint's first model
-            # (which surfaced gpt-3.5). "Default" should mean the default model.
-            if not ep_url:
-                ep_url, ep_model, ep_headers = resolve_endpoint("default", owner=user)
-            if not ep_url:
-                ep_url, ep_model, ep_headers = resolve_endpoint("chat", owner=user)
-            if not ep_url:
-                from src.database import SessionLocal
-                db = SessionLocal()
-                try:
-                    # Owner-scoped first-enabled fallback: the caller's own rows
-                    # + legacy null-owner shared rows only — never borrow another
-                    # user's private endpoint/api_key. Same fix as the
-                    # /api/v1/chat fallback (webhook_routes._first_enabled_endpoint).
-                    ep = _owned_enabled_endpoint(db, user)
-                    if ep:
-                        resolved = _resolve_endpoint_runtime(ep, owner=user)
-                        if resolved:
-                            ep_url, ep_model, ep_headers = resolved
-                finally:
-                    db.close()
-            if not ep_url:
-                raise HTTPException(400, "No endpoints configured. Add one in Settings first.")
-            if body.model:
-                ep_model = body.model
-
-        # max_rounds=0 → "Auto", let AI decide; pass 20 as the safety cap.
-        effective_max_rounds = body.max_rounds if body.max_rounds > 0 else 20
-        research_handler.start_research(
-            session_id=session_id,
-            query=body.query,
-            llm_endpoint=ep_url,
-            llm_model=ep_model,
-            max_time=body.max_time,
-            llm_headers=ep_headers,
-            max_rounds=effective_max_rounds,
-            search_provider=body.search_provider or None,
-            category=body.category or None,
-            extraction_timeout=body.extraction_timeout,
-            extraction_concurrency=body.extraction_concurrency,
-            owner=user,
-        )
-        return {"session_id": session_id, "status": "running", "query": body.query}
-
-    @router.get("/api/research/stream/{session_id}")
-    async def research_stream(session_id: str, request: Request):
-        """SSE stream of research progress events."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        if not _owns_in_memory(session_id, user):
-            raise HTTPException(404, "No research found for this session")
-        async def _generate():
-            last_progress = None
-            while True:
-                status = research_handler.get_status(session_id)
-                if status is None:
-                    yield f"data: {json.dumps({'status': 'not_found'})}\n\n"
-                    return
-                st = status.get("status", "")
-                progress = status.get("progress", {})
-                if progress != last_progress:
-                    last_progress = progress
-                    yield f"data: {json.dumps({**progress, 'status': st})}\n\n"
-                if st != "running":
-                    final = {'status': st, 'final': True}
-                    task = research_handler._active_tasks.get(session_id, {})
-                    if st == "error" and task.get("result"):
-                        final['error'] = str(task["result"])[:500]
-                    yield f"data: {json.dumps(final)}\n\n"
-                    return
-                await asyncio.sleep(1.5)
-
-        return StreamingResponse(
-            _generate(),
-            media_type="text/event-stream",
-            headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
-        )
-
-    @router.post("/api/research/result-peek/{session_id}")
-    async def research_result_peek(session_id: str, request: Request):
-        """Get research result without clearing it (for panel use)."""
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        owned_disk_path = _require_owned_or_active_research_path(session_id, user)
-        result = research_handler.get_result(session_id)
-        if result is None:
-            p = owned_disk_path
-            if p is not None:
-                d = json.loads(p.read_text(encoding="utf-8"))
-                return {
-                    "result": d.get("result", ""),
-                    "sources": d.get("sources", []),
-                    "raw_findings": d.get("raw_findings", []),
-                    "category": d.get("category") or "",
-                }
-            raise HTTPException(404, "No research result available")
-        sources = research_handler.get_sources(session_id) or []
-        raw_findings = research_handler.get_raw_findings(session_id) or []
-        return {"result": result, "sources": sources, "raw_findings": raw_findings, "category": ""}
-
-    @router.post("/api/research/spinoff/{session_id}")
-    async def research_spinoff(session_id: str, request: Request):
-        """Create a new chat session pre-seeded with this research as context.
-
-        Reads the persisted research result + sources for `session_id`, creates
-        a fresh session (inheriting endpoint/model/headers from the source
-        session if available, otherwise from the resolved chat endpoint), and
-        injects a single system message containing the report and sources so
-        the user can ask follow-up questions in a clean conversation.
-        """
-        user = _require_user(request)
-        _validate_session_id(session_id)
-        # SECURITY: gate on ownership before reading the persisted research —
-        # otherwise any authenticated user could spin off (and thereby read)
-        # another user's report by guessing its session ID. Mirrors every other
-        # endpoint in this file (see result_peek above).
-        owned_disk_path = _require_owned_or_active_research_path(session_id, user)
-        if session_manager is None:
-            raise HTTPException(500, "session_manager not configured")
-
-        # Load research data — prefer in-memory result, fall back to disk
-        result = research_handler.get_result(session_id)
-        sources = research_handler.get_sources(session_id) or []
-        query = ""
-
-        path = owned_disk_path
-        if path is not None:
-            try:
-                disk = json.loads(path.read_text(encoding="utf-8"))
-                if not result:
-                    result = disk.get("result")
-                if not sources:
-                    sources = disk.get("sources", []) or []
-                query = disk.get("query", "") or ""
-            except Exception as e:
-                logger.warning(f"Could not read research JSON for spinoff: {e}")
-
-        if not result:
-            raise HTTPException(404, "No research result available for this session")
-
-        # Inherit endpoint/model/headers from the source session when possible.
-        # For panel-launched research (rp-* IDs), there is no chat session, so
-        # fall back through the same chain as /api/research/start: research →
-        # utility → first enabled endpoint in the DB.
-        ep_url, ep_model, ep_headers = "", "", {}
-        try:
-            src_sess = session_manager.get_session(session_id)
-            ep_url = src_sess.endpoint_url or ""
-            ep_model = src_sess.model or ""
-            ep_headers = dict(src_sess.headers or {})
-        except KeyError:
-            pass
-
-        def _merge(r_url, r_model, r_headers):
-            nonlocal ep_url, ep_model, ep_headers
-            if not ep_url and r_url:
-                ep_url = r_url
-            if not ep_model and r_model:
-                ep_model = r_model
-            if not ep_headers and r_headers:
-                ep_headers = dict(r_headers)
-
-        if not ep_url or not ep_model:
-            _merge(*resolve_endpoint("chat", owner=user))
-        if not ep_url or not ep_model:
-            _merge(*resolve_endpoint("research", owner=user))
-        if not ep_url or not ep_model:
-            _merge(*resolve_endpoint("utility", owner=user))
-        if not ep_url or not ep_model:
-            # Last resort: this user's enabled endpoint, plus legacy shared rows.
-            from src.database import SessionLocal
-            from src.endpoint_resolver import normalize_base, build_chat_url, build_headers
-            db = SessionLocal()
-            try:
-                ep = _owned_enabled_endpoint(db, user)
-                if ep:
-                    base = normalize_base(ep.base_url)
-                    fallback_url = build_chat_url(base)
-                    fallback_headers = build_headers(ep.api_key, base)
-                    fallback_model = ""
-                    if ep.cached_models:
-                        try:
-                            models = json.loads(ep.cached_models)
-                            if models:
-                                fallback_model = _first_chat_model(models)
-                        except Exception:
-                            pass
-                    _merge(fallback_url, fallback_model, fallback_headers)
-            finally:
-                db.close()
-
-        if not ep_url or not ep_model:
-            raise HTTPException(400, "No endpoint configured — add one in Settings first")
-
-        # Create new session
-        new_sid = str(uuid.uuid4())
-
-        title_query = (query or "research").strip()
-        if len(title_query) > 60:
-            title_query = title_query[:57] + "…"
-        new_name = f"Follow-up: {title_query}"
-
-        new_sess = session_manager.create_session(
-            session_id=new_sid,
-            name=new_name,
-            endpoint_url=ep_url,
-            model=ep_model,
-            rag=False,
-            owner=user,
-        )
-        if ep_headers:
-            new_sess.headers = ep_headers
-            session_manager.save_sessions()
-        try:
-            from src.event_bus import fire_event
-            fire_event("session_created", user)
-        except Exception:
-            logger.debug("session_created event dispatch failed", exc_info=True)
-
-        # Build the priming system message — report only, no sources injected.
-        # The user can open the visual report for source details; keeping sources
-        # out of the chat context saves tokens and avoids the AI fabricating
-        # citations.
-        date_str = datetime.utcnow().strftime("%Y-%m-%d")
-        primer = (
-            f"[Research context — {date_str}]\n\n"
-            f"The user previously ran a deep research investigation. Use the "
-            f"report below as your primary knowledge base when answering "
-            f"follow-up questions. If the user asks something not covered, "
-            f"say so plainly rather than guessing.\n\n"
-            f"=== ORIGINAL QUERY ===\n{query or '(not recorded)'}\n\n"
-            f"=== REPORT ===\n{result}"
-        )
-
-        from core.models import ChatMessage
-        new_sess.add_message(ChatMessage(
-            role="system",
-            content=primer,
-            metadata={"research_spinoff_from": session_id},
-        ))
-        session_manager.save_sessions()
-
-        return {
-            "session_id": new_sid,
-            "name": new_name,
-            "source_count": len(sources),
-        }
-
-    return router

routes/research_routes.py

@@ -1,17 +1,678 @@
-"""Backward-compat shim — canonical location is routes/research/research_routes.py.
+"""Research background task routes — /api/research/*."""
 
-This module is replaced in ``sys.modules`` by the canonical module object so
-that ``import routes.research_routes``, ``from routes.research_routes import X``,
-``importlib.import_module("routes.research_routes")``, and
-``monkeypatch.setattr("routes.research_routes.ATTR", ...)`` (string-targeted
-patch used by ``test_research_owner_scope_routes.py``) all operate on the
-*same* object the application actually uses. Keeps existing import paths
-working after slice 2b (#4082/#4071). Source-introspection tests read the
-canonical file by path.
-"""
+import asyncio
+import json
+import logging
+import re
+import uuid
+from datetime import datetime
+from pathlib import Path
+from typing import Optional
 
-import sys as _sys
+from fastapi import APIRouter, HTTPException, Query, Request
+from fastapi.responses import HTMLResponse, StreamingResponse
+from pydantic import BaseModel, Field
+from core.middleware import INTERNAL_TOOL_USER
+from src.endpoint_resolver import resolve_endpoint
+from src.auth_helpers import _auth_disabled, get_current_user
+from core.auth import RESERVED_USERNAMES
+from src.constants import DEEP_RESEARCH_DIR
 
-from routes.research import research_routes as _canonical  # noqa: F401
+_SESSION_ID_RE = re.compile(r"^[a-zA-Z0-9-]{1,128}$")
 
-_sys.modules[__name__] = _canonical
+logger = logging.getLogger(__name__)
+
+# Model-name substrings that are NOT chat/generation models — research must
+# never pick these as its model. An OpenAI-style endpoint often lists
+# `text-embedding-ada-002` etc. first in its model list, which is why research
+# was failing with "Cannot reach model 'text-embedding-ada-002'".
+_NON_CHAT_MODEL = (
+    "text-embedding", "embedding", "tts-", "whisper", "dall-e",
+    "moderation", "rerank", "reranker", "clip", "stable-diffusion",
+)
+
+
+def _first_chat_model(models) -> str:
+    """First model that isn't an embedding/tts/etc. — falls back to models[0]."""
+    for m in (models or []):
+        if not any(p in str(m).lower() for p in _NON_CHAT_MODEL):
+            return m
+    return (models[0] if models else "")
+
+
+def _resolve_research_endpoint(sess, owner: Optional[str] = None) -> tuple:
+    """Return (endpoint_url, model, headers) for Deep Research, checking admin overrides."""
+    owner = owner or getattr(sess, "owner", None) or None
+    url, model, headers = resolve_endpoint(
+        "research",
+        fallback_url=sess.endpoint_url,
+        fallback_model=sess.model,
+        fallback_headers=sess.headers,
+        owner=owner,
+    )
+    return url, model, headers
+
+
+def _owned_enabled_endpoint(db, owner, endpoint_id=None):
+    """An enabled ModelEndpoint VISIBLE to `owner` (their own rows + legacy
+    null-owner "shared" rows), optionally narrowed to a specific endpoint_id;
+    None if nothing visible matches.
+
+    Owner-scoped on purpose. ModelEndpoint is per-user (core/database.py: non-null
+    owner = private, "the model picker only shows the endpoint to that user") and
+    holds a decrypted `api_key`. /api/research/start feeds the resolved row's
+    api_key + base_url into research_handler.start_research(llm_endpoint=,
+    llm_headers=), so an UNSCOPED lookup — by the caller-supplied endpoint_id, or
+    via the bare first-enabled fallback — would let a research-privileged user
+    spend ANOTHER user's API key/quota and reach whatever internal base_url they
+    configured. Mirrors webhook_routes._first_enabled_endpoint and
+    session_routes._owned_endpoint. A null/empty owner is a no-op (single-user /
+    legacy mode).
+    """
+    from src.database import ModelEndpoint
+    from src.auth_helpers import owner_filter
+    q = db.query(ModelEndpoint).filter(ModelEndpoint.is_enabled == True)  # noqa: E712
+    if endpoint_id:
+        q = q.filter(ModelEndpoint.id == endpoint_id)
+    return owner_filter(q, ModelEndpoint, owner).first()
+
+
+def _resolve_endpoint_runtime(ep, owner=None, model: Optional[str] = None):
+    """Resolve a ModelEndpoint row into (chat_url, model, headers).
+
+    Mirrors endpoint_resolver.resolve_endpoint's provider-auth handling for
+    panel-selected research endpoints. ChatGPT Subscription endpoints keep
+    OAuth tokens in ProviderAuthSession, so ep.api_key is intentionally empty.
+    """
+    from src.endpoint_resolver import (
+        build_chat_url,
+        build_headers,
+        resolve_endpoint_runtime as resolve_model_endpoint_runtime,
+    )
+
+    try:
+        base, api_key = resolve_model_endpoint_runtime(ep, owner=owner)
+    except Exception as e:
+        logger.warning("Could not resolve endpoint credentials for research: %s", e)
+        return None
+
+    ep_model = (model or "").strip()
+    if not ep_model:
+        try:
+            models = json.loads(ep.cached_models) if ep.cached_models else []
+            if models:
+                ep_model = _first_chat_model(models)
+        except Exception:
+            pass
+    if not ep_model:
+        return None
+    return build_chat_url(base), ep_model, build_headers(api_key, base)
+
+
+def setup_research_routes(research_handler, session_manager=None) -> APIRouter:
+    router = APIRouter(tags=["research"])
+
+    def _require_user(request: Request) -> str:
+        """All research endpoints require an authenticated user. Research
+        data isn't owner-scoped in the on-disk JSON yet, so we at least
+        block anonymous access. Multi-tenant deploys should additionally
+        verify the session belongs to this user."""
+        user = get_current_user(request)
+        if not user:
+            if _auth_disabled():
+                return ""
+            raise HTTPException(401, "Not authenticated")
+        return user
+
+    def _validate_session_id(session_id: str) -> None:
+        if not _SESSION_ID_RE.fullmatch(session_id):
+            raise HTTPException(400, "Invalid session ID format")
+
+    def _owns_in_memory(session_id: str, user: str) -> bool:
+        """Ownership check for an in-flight (in-memory) research task.
+        Falls back to the on-disk JSON if the task has already finished."""
+        entry = research_handler._active_tasks.get(session_id)
+        if entry is not None:
+            return entry.get("owner", "") == user
+        # Task no longer in memory — check the persisted JSON.
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
+        if not path.exists():
+            return False
+        try:
+            return json.loads(path.read_text(encoding="utf-8")).get("owner") == user
+        except Exception:
+            return False
+
+    @router.get("/api/research/active")
+    async def research_active(request: Request):
+        """List all currently active (running) research tasks."""
+        user = _require_user(request)
+        active = []
+        for sid, entry in research_handler._active_tasks.items():
+            # SECURITY: only show this user's running tasks.
+            if entry.get("owner", "") != user:
+                continue
+            if entry.get("status") == "running":
+                active.append({
+                    "session_id": sid,
+                    "query": entry.get("query", ""),
+                    "status": "running",
+                    "progress": entry.get("progress", {}),
+                    "started_at": entry.get("started_at", 0),
+                })
+        return {"active": active}
+
+    @router.get("/api/research/status/{session_id}")
+    async def research_status(session_id: str, request: Request):
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        if not _owns_in_memory(session_id, user):
+            raise HTTPException(404, "No research found for this session")
+        status = research_handler.get_status(session_id)
+        if status is None:
+            raise HTTPException(404, "No research found for this session")
+        return status
+
+    @router.post("/api/research/cancel/{session_id}")
+    async def research_cancel(session_id: str, request: Request):
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        if not _owns_in_memory(session_id, user):
+            raise HTTPException(404, "No research found for this session")
+        cancelled = research_handler.cancel_research(session_id)
+        return {"cancelled": cancelled}
+
+    @router.post("/api/research/result/{session_id}")
+    async def research_result(session_id: str, request: Request):
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        if not _owns_in_memory(session_id, user):
+            raise HTTPException(404, "No research result available")
+        result = research_handler.get_result(session_id)
+        if result is None:
+            raise HTTPException(404, "No research result available")
+        sources = research_handler.get_sources(session_id) or []
+        raw_findings = research_handler.get_raw_findings(session_id) or []
+        research_handler.clear_result(session_id)
+        return {"result": result, "sources": sources, "raw_findings": raw_findings}
+
+    def _assert_owns_research(session_id: str, user: str) -> None:
+        """404-not-403 ownership gate for a research session's on-disk JSON.
+        Use BEFORE returning any data or mutating the file."""
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
+        if not path.exists():
+            raise HTTPException(404, "Research not found")
+        try:
+            owner = json.loads(path.read_text(encoding="utf-8")).get("owner")
+        except Exception:
+            raise HTTPException(404, "Research not found")
+        if owner != user:
+            raise HTTPException(404, "Research not found")
+
+    @router.get("/api/research/report/{session_id}")
+    async def research_report(session_id: str, request: Request):
+        """Serve the visual HTML report for a completed research session."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        _assert_owns_research(session_id, user)
+        logger.info(f"Visual report requested for session {session_id}")
+        try:
+            html_content = research_handler.get_report_html(session_id)
+        except Exception as e:
+            logger.error(f"Visual report generation error: {e}", exc_info=True)
+            raise HTTPException(500, f"Report generation failed: {e}")
+        if html_content is None:
+            logger.warning(f"No report data found for session {session_id}")
+            raise HTTPException(404, "No visual report available for this session")
+        return HTMLResponse(content=html_content)
+
+    class HideImageRequest(BaseModel):
+        url: str
+
+    @router.post("/api/research/{session_id}/hide-image")
+    async def research_hide_image(session_id: str, body: HideImageRequest, request: Request):
+        """Mark an image URL as hidden for this research's visual report.
+        Persisted to the research JSON so subsequent /report renders skip it."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        _assert_owns_research(session_id, user)
+        ok = research_handler.hide_image(session_id, body.url)
+        if not ok:
+            raise HTTPException(404, "Research not found")
+        return {"ok": True}
+
+    @router.post("/api/research/{session_id}/unhide-images")
+    async def research_unhide_images(session_id: str, request: Request):
+        """Clear the hidden-images list for a research session."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        _assert_owns_research(session_id, user)
+        ok = research_handler.unhide_all_images(session_id)
+        if not ok:
+            raise HTTPException(404, "Research not found")
+        return {"ok": True}
+
+    @router.get("/api/research/library")
+    async def research_library(
+        request: Request,
+        search: Optional[str] = Query(None),
+        sort: str = Query("recent"),
+        limit: int = Query(50),
+        archived: bool = Query(False),
+    ):
+        user = _require_user(request)
+        """List all completed research for the Library panel."""
+        data_dir = Path(DEEP_RESEARCH_DIR)
+        items = []
+        for p in data_dir.glob("*.json"):
+            try:
+                d = json.loads(p.read_text(encoding="utf-8"))
+                # SECURITY: only show research belonging to this user. Legacy
+                # JSONs without an `owner` field are hidden — auth was the only
+                # gate before, so every user saw every other user's reports.
+                if d.get("owner") != user:
+                    continue
+                # Archived view shows ONLY archived reports; default hides them.
+                if bool(d.get("archived")) != archived:
+                    continue
+                query = d.get("query", "")
+                if search and search.lower() not in query.lower():
+                    continue
+                sources = d.get("sources", [])
+                items.append({
+                    "id": p.stem,
+                    "query": query,
+                    "category": d.get("category") or "",
+                    "source_count": len(sources),
+                    "status": d.get("status", "done"),
+                    "duration": d.get("stats", {}).get("Duration", ""),
+                    "rounds": d.get("stats", {}).get("Rounds", ""),
+                    "started_at": d.get("started_at", 0),
+                    "completed_at": d.get("completed_at", 0),
+                    "archived": bool(d.get("archived")),
+                })
+            except Exception:
+                continue
+
+        # Sort
+        if sort == "recent":
+            items.sort(key=lambda x: x["completed_at"] or 0, reverse=True)
+        elif sort == "oldest":
+            items.sort(key=lambda x: x["completed_at"] or 0)
+        elif sort == "most-messages":
+            items.sort(key=lambda x: x["source_count"], reverse=True)
+        elif sort == "alpha":
+            items.sort(key=lambda x: x["query"].lower())
+
+        return {"research": items[:limit], "total": len(items)}
+
+    @router.get("/api/research/detail/{session_id}")
+    async def research_detail(session_id: str, request: Request):
+        """Return the full JSON for a single research result — sources,
+        summary, stats — used by the Library preview panel."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
+        if not path.exists():
+            raise HTTPException(404, "Research not found")
+        try:
+            data = json.loads(path.read_text(encoding="utf-8"))
+        except Exception as e:
+            raise HTTPException(500, f"Failed to read research: {e}")
+        # SECURITY: 404 (not 403) so we don't leak that the report exists.
+        if data.get("owner") != user:
+            raise HTTPException(404, "Research not found")
+        return data
+
+    @router.post("/api/research/{session_id}/archive")
+    async def research_archive(session_id: str, request: Request, archived: bool = Query(True)):
+        """Soft-archive / restore a research report (sets `archived` in its JSON)."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
+        if not path.exists():
+            raise HTTPException(404, "Research not found")
+        try:
+            data = json.loads(path.read_text(encoding="utf-8"))
+            if data.get("owner") != user:
+                raise HTTPException(404, "Research not found")
+            data["archived"] = bool(archived)
+            path.write_text(json.dumps(data), encoding="utf-8")
+        except HTTPException:
+            raise
+        except Exception as e:
+            raise HTTPException(500, f"Failed to update research: {e}")
+        return {"ok": True, "id": session_id, "archived": bool(archived)}
+
+    @router.delete("/api/research/{session_id}")
+    async def research_delete(session_id: str, request: Request):
+        """Delete a research result from disk."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        data_dir = Path(DEEP_RESEARCH_DIR)
+        json_path = data_dir / f"{session_id}.json"
+        deleted = False
+        if json_path.exists():
+            # SECURITY: verify ownership before letting the caller delete it.
+            try:
+                data = json.loads(json_path.read_text(encoding="utf-8"))
+                if data.get("owner") != user:
+                    raise HTTPException(404, "Research not found")
+            except HTTPException:
+                raise
+            except Exception:
+                raise HTTPException(404, "Research not found")
+            json_path.unlink()
+            deleted = True
+        return {"deleted": deleted}
+
+    # ------------------------------------------------------------------
+    # Panel endpoints — launch research without a chat session
+    # ------------------------------------------------------------------
+
+    class ResearchStartRequest(BaseModel):
+        query: str
+        # max_rounds=0 means "Auto" — let the AI decide when to stop, capped at 20.
+        max_rounds: int = Field(default=0, ge=0, le=20)
+        search_provider: Optional[str] = None
+        endpoint_id: Optional[str] = None
+        model: Optional[str] = None
+        max_time: int = Field(default=300, ge=60, le=1800)
+        extraction_timeout: Optional[int] = Field(default=None, ge=15, le=3600)
+        extraction_concurrency: Optional[int] = Field(default=None, ge=1, le=12)
+        category: Optional[str] = None
+
+    @router.post("/api/research/start")
+    async def research_start(body: ResearchStartRequest, request: Request):
+        """Launch a research job from the dedicated panel."""
+        from src.auth_helpers import require_privilege
+        user = require_privilege(request, "can_use_research")
+        if user == INTERNAL_TOOL_USER:
+            tool_owner = (request.headers.get("X-Odysseus-Owner") or "").strip()
+            if tool_owner and tool_owner not in RESERVED_USERNAMES:
+                auth_mgr = getattr(request.app.state, "auth_manager", None)
+                if auth_mgr is not None and getattr(auth_mgr, "is_configured", False):
+                    try:
+                        privs = auth_mgr.get_privileges(tool_owner) or {}
+                        if not privs.get("can_use_research", True):
+                            raise HTTPException(403, f"Your account is not allowed to can use research.")
+                    except HTTPException:
+                        raise
+                    except Exception:
+                        pass
+                user = tool_owner
+        session_id = f"rp-{uuid.uuid4().hex[:12]}"
+
+        if body.endpoint_id:
+            from src.database import SessionLocal
+            db = SessionLocal()
+            try:
+                # Owner-scoped: never resolve another user's private endpoint
+                # (and its decrypted api_key / internal base_url). A scoped miss
+                # reads as 404 so the endpoint's existence isn't revealed.
+                ep = _owned_enabled_endpoint(db, user, body.endpoint_id)
+                if not ep:
+                    raise HTTPException(404, "Endpoint not found or disabled")
+                resolved = _resolve_endpoint_runtime(ep, owner=user, model=body.model)
+                if not resolved:
+                    raise HTTPException(400, "Endpoint is not configured with a usable model.")
+                ep_url, ep_model, ep_headers = resolved
+            finally:
+                db.close()
+        else:
+            ep_url, ep_model, ep_headers = resolve_endpoint("research", owner=user)
+            if not ep_url:
+                ep_url, ep_model, ep_headers = resolve_endpoint("utility", owner=user)
+            # When neither research nor utility is configured, use the user's
+            # configured DEFAULT model (default_endpoint_id/default_model) rather
+            # than arbitrarily grabbing the first enabled endpoint's first model
+            # (which surfaced gpt-3.5). "Default" should mean the default model.
+            if not ep_url:
+                ep_url, ep_model, ep_headers = resolve_endpoint("default", owner=user)
+            if not ep_url:
+                ep_url, ep_model, ep_headers = resolve_endpoint("chat", owner=user)
+            if not ep_url:
+                from src.database import SessionLocal
+                db = SessionLocal()
+                try:
+                    # Owner-scoped first-enabled fallback: the caller's own rows
+                    # + legacy null-owner shared rows only — never borrow another
+                    # user's private endpoint/api_key. Same fix as the
+                    # /api/v1/chat fallback (webhook_routes._first_enabled_endpoint).
+                    ep = _owned_enabled_endpoint(db, user)
+                    if ep:
+                        resolved = _resolve_endpoint_runtime(ep, owner=user)
+                        if resolved:
+                            ep_url, ep_model, ep_headers = resolved
+                finally:
+                    db.close()
+            if not ep_url:
+                raise HTTPException(400, "No endpoints configured. Add one in Settings first.")
+            if body.model:
+                ep_model = body.model
+
+        # max_rounds=0 → "Auto", let AI decide; pass 20 as the safety cap.
+        effective_max_rounds = body.max_rounds if body.max_rounds > 0 else 20
+        research_handler.start_research(
+            session_id=session_id,
+            query=body.query,
+            llm_endpoint=ep_url,
+            llm_model=ep_model,
+            max_time=body.max_time,
+            llm_headers=ep_headers,
+            max_rounds=effective_max_rounds,
+            search_provider=body.search_provider or None,
+            category=body.category or None,
+            extraction_timeout=body.extraction_timeout,
+            extraction_concurrency=body.extraction_concurrency,
+            owner=user,
+        )
+        return {"session_id": session_id, "status": "running", "query": body.query}
+
+    @router.get("/api/research/stream/{session_id}")
+    async def research_stream(session_id: str, request: Request):
+        """SSE stream of research progress events."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        if not _owns_in_memory(session_id, user):
+            raise HTTPException(404, "No research found for this session")
+        async def _generate():
+            last_progress = None
+            while True:
+                status = research_handler.get_status(session_id)
+                if status is None:
+                    yield f"data: {json.dumps({'status': 'not_found'})}\n\n"
+                    return
+                st = status.get("status", "")
+                progress = status.get("progress", {})
+                if progress != last_progress:
+                    last_progress = progress
+                    yield f"data: {json.dumps({**progress, 'status': st})}\n\n"
+                if st != "running":
+                    final = {'status': st, 'final': True}
+                    task = research_handler._active_tasks.get(session_id, {})
+                    if st == "error" and task.get("result"):
+                        final['error'] = str(task["result"])[:500]
+                    yield f"data: {json.dumps(final)}\n\n"
+                    return
+                await asyncio.sleep(1.5)
+
+        return StreamingResponse(
+            _generate(),
+            media_type="text/event-stream",
+            headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
+        )
+
+    @router.post("/api/research/result-peek/{session_id}")
+    async def research_result_peek(session_id: str, request: Request):
+        """Get research result without clearing it (for panel use)."""
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        if not _owns_in_memory(session_id, user):
+            raise HTTPException(404, "No research found for this session")
+        result = research_handler.get_result(session_id)
+        if result is None:
+            p = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
+            if p.exists():
+                d = json.loads(p.read_text(encoding="utf-8"))
+                return {
+                    "result": d.get("result", ""),
+                    "sources": d.get("sources", []),
+                    "raw_findings": d.get("raw_findings", []),
+                    "category": d.get("category") or "",
+                }
+            raise HTTPException(404, "No research result available")
+        sources = research_handler.get_sources(session_id) or []
+        raw_findings = research_handler.get_raw_findings(session_id) or []
+        return {"result": result, "sources": sources, "raw_findings": raw_findings, "category": ""}
+
+    @router.post("/api/research/spinoff/{session_id}")
+    async def research_spinoff(session_id: str, request: Request):
+        """Create a new chat session pre-seeded with this research as context.
+
+        Reads the persisted research result + sources for `session_id`, creates
+        a fresh session (inheriting endpoint/model/headers from the source
+        session if available, otherwise from the resolved chat endpoint), and
+        injects a single system message containing the report and sources so
+        the user can ask follow-up questions in a clean conversation.
+        """
+        user = _require_user(request)
+        _validate_session_id(session_id)
+        # SECURITY: gate on ownership before reading the persisted research —
+        # otherwise any authenticated user could spin off (and thereby read)
+        # another user's report by guessing its session ID. Mirrors every other
+        # endpoint in this file (see result_peek above).
+        if not _owns_in_memory(session_id, user):
+            raise HTTPException(404, "No research found for this session")
+        if session_manager is None:
+            raise HTTPException(500, "session_manager not configured")
+
+        # Load research data — prefer in-memory result, fall back to disk
+        result = research_handler.get_result(session_id)
+        sources = research_handler.get_sources(session_id) or []
+        query = ""
+
+        path = Path(DEEP_RESEARCH_DIR) / f"{session_id}.json"
+        if path.exists():
+            try:
+                disk = json.loads(path.read_text(encoding="utf-8"))
+                if not result:
+                    result = disk.get("result")
+                if not sources:
+                    sources = disk.get("sources", []) or []
+                query = disk.get("query", "") or ""
+            except Exception as e:
+                logger.warning(f"Could not read research JSON for spinoff: {e}")
+
+        if not result:
+            raise HTTPException(404, "No research result available for this session")
+
+        # Inherit endpoint/model/headers from the source session when possible.
+        # For panel-launched research (rp-* IDs), there is no chat session, so
+        # fall back through the same chain as /api/research/start: research →
+        # utility → first enabled endpoint in the DB.
+        ep_url, ep_model, ep_headers = "", "", {}
+        try:
+            src_sess = session_manager.get_session(session_id)
+            ep_url = src_sess.endpoint_url or ""
+            ep_model = src_sess.model or ""
+            ep_headers = dict(src_sess.headers or {})
+        except KeyError:
+            pass
+
+        def _merge(r_url, r_model, r_headers):
+            nonlocal ep_url, ep_model, ep_headers
+            if not ep_url and r_url:
+                ep_url = r_url
+            if not ep_model and r_model:
+                ep_model = r_model
+            if not ep_headers and r_headers:
+                ep_headers = dict(r_headers)
+
+        if not ep_url or not ep_model:
+            _merge(*resolve_endpoint("chat", owner=user))
+        if not ep_url or not ep_model:
+            _merge(*resolve_endpoint("research", owner=user))
+        if not ep_url or not ep_model:
+            _merge(*resolve_endpoint("utility", owner=user))
+        if not ep_url or not ep_model:
+            # Last resort: this user's enabled endpoint, plus legacy shared rows.
+            from src.database import SessionLocal
+            from src.endpoint_resolver import normalize_base, build_chat_url, build_headers
+            db = SessionLocal()
+            try:
+                ep = _owned_enabled_endpoint(db, user)
+                if ep:
+                    base = normalize_base(ep.base_url)
+                    fallback_url = build_chat_url(base)
+                    fallback_headers = build_headers(ep.api_key, base)
+                    fallback_model = ""
+                    if ep.cached_models:
+                        try:
+                            models = json.loads(ep.cached_models)
+                            if models:
+                                fallback_model = _first_chat_model(models)
+                        except Exception:
+                            pass
+                    _merge(fallback_url, fallback_model, fallback_headers)
+            finally:
+                db.close()
+
+        if not ep_url or not ep_model:
+            raise HTTPException(400, "No endpoint configured — add one in Settings first")
+
+        # Create new session
+        new_sid = str(uuid.uuid4())
+
+        title_query = (query or "research").strip()
+        if len(title_query) > 60:
+            title_query = title_query[:57] + "…"
+        new_name = f"Follow-up: {title_query}"
+
+        new_sess = session_manager.create_session(
+            session_id=new_sid,
+            name=new_name,
+            endpoint_url=ep_url,
+            model=ep_model,
+            rag=False,
+            owner=user,
+        )
+        if ep_headers:
+            new_sess.headers = ep_headers
+            session_manager.save_sessions()
+        try:
+            from src.event_bus import fire_event
+            fire_event("session_created", user)
+        except Exception:
+            logger.debug("session_created event dispatch failed", exc_info=True)
+
+        # Build the priming system message — report only, no sources injected.
+        # The user can open the visual report for source details; keeping sources
+        # out of the chat context saves tokens and avoids the AI fabricating
+        # citations.
+        date_str = datetime.utcnow().strftime("%Y-%m-%d")
+        primer = (
+            f"[Research context — {date_str}]\n\n"
+            f"The user previously ran a deep research investigation. Use the "
+            f"report below as your primary knowledge base when answering "
+            f"follow-up questions. If the user asks something not covered, "
+            f"say so plainly rather than guessing.\n\n"
+            f"=== ORIGINAL QUERY ===\n{query or '(not recorded)'}\n\n"
+            f"=== REPORT ===\n{result}"
+        )
+
+        from core.models import ChatMessage
+        new_sess.add_message(ChatMessage(
+            role="system",
+            content=primer,
+            metadata={"research_spinoff_from": session_id},
+        ))
+        session_manager.save_sessions()
+
+        return {
+            "session_id": new_sid,
+            "name": new_name,
+            "source_count": len(sources),
+        }
+
+    return router

routes/session_routes.py

@@ -162,7 +162,7 @@ def _persist_session_headers(session_id: str, headers: dict | None) -> None:
         db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
         if db_session:
             db_session.headers = headers or {}
-            db_session.updated_at = utcnow_naive()
+            db_session.updated_at = datetime.utcnow()
             db.commit()
     except Exception:
         db.rollback()
@@ -207,7 +207,6 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
     """Setup session routes with the provided manager and config"""
 
     REQUEST_TIMEOUT = config.get("REQUEST_TIMEOUT", 20)
-    SESSION_MODEL_VALIDATION_TIMEOUT = min(float(REQUEST_TIMEOUT or 20), 3.0)
     OPENAI_API_KEY = config.get("OPENAI_API_KEY")
     SESSIONS_FILE = config.get("SESSIONS_FILE")
     
@@ -224,8 +223,8 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
         # purge exists only to catch ghosts the frontend missed (tab close,
         # crash). Only clean up rows old enough to be definitely orphaned.
         try:
-            from datetime import timedelta as _td
-            _cutoff = utcnow_naive() - _td(minutes=10)
+            from datetime import datetime as _dt, timedelta as _td
+            _cutoff = _dt.utcnow() - _td(minutes=10)
             _purge_db = SessionLocal()
             try:
                 from core.database import ChatMessage as _DbMsg
@@ -375,7 +374,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             from src.llm_core import list_model_ids
             ids = list_model_ids(
                 endpoint_url,
-                timeout=SESSION_MODEL_VALIDATION_TIMEOUT,
+                timeout=REQUEST_TIMEOUT,
                 headers=validation_headers,
                 owner=user,
                 endpoint_id=endpoint_id.strip() if endpoint_id else None,
@@ -395,7 +394,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             req_base = _os.path.basename(model_to_use.rstrip("/"))
             avail = list_model_ids(
                 endpoint_url,
-                timeout=SESSION_MODEL_VALIDATION_TIMEOUT,
+                timeout=REQUEST_TIMEOUT,
                 headers=validation_headers,
                 owner=user,
                 endpoint_id=endpoint_id.strip() if endpoint_id else None,
@@ -471,7 +470,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                 db_session = db.query(DbSession).filter(DbSession.id == sid).first()
                 if db_session:
                     db_session.folder = folder if folder else None
-                    db_session.updated_at = utcnow_naive()
+                    db_session.updated_at = datetime.utcnow()
                     db.commit()
                     result["folder"] = folder if folder else None
             finally:
@@ -518,7 +517,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                     db_session.model = model
                     db_session.endpoint_url = endpoint_url
                     db_session.headers = session.headers or {}
-                    db_session.updated_at = utcnow_naive()
+                    db_session.updated_at = datetime.utcnow()
                     db.commit()
             finally:
                 db.close()
@@ -647,7 +646,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                 db_session = db.query(DbSession).filter(DbSession.id == sid).first()
                 if db_session:
                     db_session.archived = True
-                    db_session.updated_at = utcnow_naive()
+                    db_session.updated_at = datetime.utcnow()
                     db.commit()
                     
                     # Update in memory if it exists
@@ -681,7 +680,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             if not db_session:
                 raise HTTPException(404, f"Session {sid} not found")
             db_session.archived = False
-            db_session.updated_at = utcnow_naive()
+            db_session.updated_at = datetime.utcnow()
             db.commit()
             # Reload into session manager so it appears in the active list
             try:
@@ -891,7 +890,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                 db_session = db.query(DbSession).filter(DbSession.id == session_id).first()
                 if db_session:
                     db_session.is_important = important
-                    db_session.updated_at = utcnow_naive()
+                    db_session.updated_at = datetime.utcnow()
                     db.commit()
 
                     # Update in memory if it exists
@@ -980,7 +979,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
             metadata={
                 "compacted": True,
                 "summarized_count": len(older),
-                "timestamp": utcnow_naive().isoformat(),
+                "timestamp": datetime.utcnow().isoformat(),
             },
         )
         new_history = [summary_msg] + recent
@@ -1257,7 +1256,7 @@ def setup_session_routes(session_manager: SessionManager, config: dict, webhook_
                 db_session = db_session_q.first()
                 if db_session:
                     db_session.folder = folder_name
-                    db_session.updated_at = utcnow_naive()
+                    db_session.updated_at = datetime.utcnow()
                     updated += 1
             db.commit()
         except Exception as e:

routes/shell_routes.py

@@ -16,11 +16,6 @@ from pathlib import Path
 from typing import Dict, Any
 from core.platform_compat import IS_APPLE_SILICON, which_tool
 from core.middleware import INTERNAL_TOOL_USER
-from src.host_docker_access import (
-    HOST_DOCKER_ACCESS_HINT,
-    host_docker_access_enabled as _host_docker_access_enabled,
-    running_in_container as _running_in_container,
-)
 from src.optional_deps import prepare_optional_dependency_import
 
 # POSIX-only: `pty`/`fcntl` transitively import `termios`, which does NOT exist
@@ -108,17 +103,32 @@ logger = logging.getLogger(__name__)
 PTY_SUPPORTED = pty is not None and fcntl is not None and hasattr(os, "setsid")
 
 
-DOCKER_IN_CONTAINER_HINT = HOST_DOCKER_ACCESS_HINT
+DOCKER_IN_CONTAINER_HINT = (
+    "Not available inside the Odysseus container by design. The image ships no "
+    "docker CLI and no host socket is mounted. Run Docker-backed launches on a "
+    "remote server, where docker is checked over SSH. Mounting /var/run/docker.sock "
+    "into the container would grant it host-root access, so only do that if you "
+    "accept that risk."
+)
+
+
+def _running_in_container(dockerenv_path="/.dockerenv", cgroup_path="/proc/1/cgroup"):
+    if os.path.exists(dockerenv_path):
+        return True
+    try:
+        with open(cgroup_path, "r", encoding="utf-8") as fh:
+            contents = fh.read()
+    except OSError:
+        return False
+    return any(token in contents for token in ("docker", "containerd", "kubepods"))
 
 
 DockerRowStatus = namedtuple("DockerRowStatus", ["applicable", "install_hint"])
 PackageUpdateStatus = namedtuple("PackageUpdateStatus", ["available", "note"])
 
 
-def _docker_row_status(
-    *, on_remote, in_container, installed, default_hint, host_docker_access=False
-):
-    local_docker_unavailable = not on_remote and in_container and not host_docker_access
+def _docker_row_status(*, on_remote, in_container, installed, default_hint):
+    local_docker_unavailable = not on_remote and in_container and not installed
     if local_docker_unavailable:
         return DockerRowStatus(applicable=False, install_hint=DOCKER_IN_CONTAINER_HINT)
     return DockerRowStatus(applicable=True, install_hint=default_hint)
@@ -164,8 +174,6 @@ def _package_installed_from_probe(name: str, probe: dict) -> bool:
         return bool(binaries.get("llama-server") or dists.get("llama-cpp-python"))
     if name == "sglang":
         return bool(dists.get("sglang") or modules.get("sglang", {}).get("real_module"))
-    if name == "mlx_lm":
-        return bool(dists.get("mlx-lm") or modules.get("mlx_lm", {}).get("real_module"))
     if name == "diffusers":
         return bool(
             (dists.get("diffusers") or modules.get("diffusers", {}).get("real_module"))
@@ -212,10 +220,6 @@ def _package_status_note(name: str, probe: dict) -> str:
         if _package_installed_from_probe(name, probe):
             return f"diffusers {dists.get('diffusers', 'available')} with torch {dists.get('torch', 'available')}"
         return "Diffusers serving needs both diffusers and torch."
-    if name == "mlx_lm":
-        if _package_installed_from_probe(name, probe):
-            return f"MLX LM {dists.get('mlx-lm', 'available')}"
-        return "MLX serving needs mlx-lm on an Apple Silicon Mac."
     if name in dists:
         return f"{name} {dists[name]}"
     return ""
@@ -313,14 +317,12 @@ dist_names={{
     'vllm':['vllm'],
     'llama_cpp':['llama-cpp-python'],
     'sglang':['sglang'],
-    'mlx_lm':['mlx-lm'],
     'diffusers':['diffusers','torch'],
     'hf_transfer':['hf-transfer','hf_transfer'],
 }}
 bin_names={{
     'vllm':['vllm'],
     'llama_cpp':['llama-server'],
-    'tmux':['tmux'],
 }}
 
 def add_user_install_bins_to_path():
@@ -333,8 +335,6 @@ def add_user_install_bins_to_path():
     candidates.append(os.path.expanduser('~/llama.cpp/build/bin'))
     candidates.append(os.path.expanduser('~/llama.cpp/build-vulkan/bin'))
     candidates.append(os.path.expanduser('~/.local/bin'))
-    candidates.append('/opt/homebrew/bin')
-    candidates.append('/usr/local/bin')
     parts = os.environ.get('PATH', '').split(os.pathsep) if os.environ.get('PATH') else []
     changed = False
     for path in reversed([p for p in candidates if p]):
@@ -409,47 +409,6 @@ class ShellExecRequest(BaseModel):
     use_tmux: bool = False  # run in tmux session (survives browser disconnect)
 
 
-_REMOTE_TMUX_PATH_PREFIX = 'PATH="$HOME/.local/bin:$HOME/bin:/opt/homebrew/bin:/usr/local/bin:$PATH"; '
-
-
-def _normalize_legacy_remote_tmux_exec(command: str) -> str:
-    """Repair stale frontend Cookbook tmux SSH commands.
-
-    Older loaded JS sends `ssh host 'tmux capture-pane ...'`. On macOS/Homebrew
-    remotes, non-login SSH shells often lack /opt/homebrew/bin, so tmux is
-    installed but the capture/kill command returns nothing. Keep this narrowly
-    scoped to SSH commands whose remote shell starts with `tmux `.
-    """
-    cmd = command or ""
-    if _REMOTE_TMUX_PATH_PREFIX in cmd or not cmd.lstrip().startswith("ssh "):
-        return cmd
-    try:
-        parts = shlex.split(cmd)
-    except Exception:
-        return cmd
-    if not parts or parts[0] != "ssh":
-        return cmd
-    remote_idx = -1
-    i = 1
-    while i < len(parts):
-        part = parts[i]
-        if part in {"-p", "-o", "-i", "-F", "-J", "-l", "-S", "-W", "-b", "-c", "-m"}:
-            i += 2
-            continue
-        if part.startswith("-"):
-            i += 1
-            continue
-        remote_idx = i
-        break
-    if remote_idx < 0 or remote_idx + 1 >= len(parts):
-        return cmd
-    remote_cmd = " ".join(parts[remote_idx + 1:]).strip()
-    if not remote_cmd.startswith("tmux "):
-        return cmd
-    repaired = parts[:remote_idx + 1] + [_REMOTE_TMUX_PATH_PREFIX + remote_cmd]
-    return shlex.join(repaired)
-
-
 async def _create_shell(command: str, **kwargs):
     """Spawn a shell subprocess for `command`.
 
@@ -866,10 +825,6 @@ def setup_shell_routes() -> APIRouter:
         if not cmd:
             return {"stdout": "", "stderr": "No command provided", "exit_code": 1}
 
-        fixed_cmd = _normalize_legacy_remote_tmux_exec(cmd)
-        if fixed_cmd != cmd:
-            logger.info("Rewrote legacy remote tmux exec command with Homebrew PATH")
-            cmd = fixed_cmd
         logger.info("User shell exec requested: length=%d", len(cmd))
         result = await _exec_shell(
             cmd, timeout=req.timeout if req.timeout is not None else EXEC_TIMEOUT
@@ -1108,19 +1063,8 @@ def setup_shell_routes() -> APIRouter:
         importlib.invalidate_caches()
         try:
             user_site = site.getusersitepackages()
-            if user_site and os.path.isdir(user_site):
-                # Use addsitedir(), NOT a bare sys.path.append(). When a package
-                # is `pip install --user`'d at runtime (Cookbook → Install) the
-                # long-lived server process started before the user-site existed,
-                # so site never processed it — including its `.pth` hooks. On
-                # Python 3.12+ `distutils` is gone from stdlib and is only
-                # restored by setuptools' `distutils-precedence.pth`, which ships
-                # in user-site. basicsr (a realesrgan dep) does `import distutils`
-                # at import time, so a plain append left the package importable
-                # but `import distutils` failing → realesrgan probed as
-                # not-installed until a full process restart. addsitedir() replays
-                # the `.pth` files so the shim is active.
-                site.addsitedir(user_site)
+            if user_site and os.path.isdir(user_site) and user_site not in sys.path:
+                sys.path.append(user_site)
         except Exception:
             pass
         if ssh_port and str(ssh_port).strip() not in ("", "22"):
@@ -1189,13 +1133,6 @@ def setup_shell_routes() -> APIRouter:
                 "category": "LLM",
                 "target": "remote",
             },
-            {
-                "name": "mlx_lm",
-                "pip": "mlx-lm",
-                "desc": "Serve MLX-format models on Apple Silicon Macs",
-                "category": "LLM",
-                "target": "remote",
-            },
             {
                 "name": "APFEL",
                 "pip": "",
@@ -1211,7 +1148,7 @@ def setup_shell_routes() -> APIRouter:
             {
                 "name": "diffusers",
                 "pip": "diffusers[torch]",
-                "desc": "Image generation/editing pipelines (SD, Flux) with PyTorch",
+                "desc": "Image generation pipelines (SD, Flux) with PyTorch",
                 "category": "Image",
                 "target": "remote",
             },
@@ -1341,9 +1278,9 @@ def setup_shell_routes() -> APIRouter:
                 for name in all_system_names:
                     qn = shlex.quote(name)
                     checks.append(
-                        f"PATH=\"$HOME/.local/bin:$HOME/bin:/opt/homebrew/bin:/usr/local/bin:$PATH\"; if command -v {qn} >/dev/null 2>&1; then echo {qn}=1; else echo {qn}=0; fi"
+                        f"if command -v {qn} >/dev/null 2>&1; then echo {qn}=1; else echo {qn}=0; fi"
                     )
-                checks.append("echo '---OSREL---'; cat /etc/os-release 2>/dev/null || { [ \"$(uname -s 2>/dev/null)\" = \"Darwin\" ] && echo ID=macos; } || true")
+                checks.append("echo '---OSREL---'; cat /etc/os-release 2>/dev/null || true")
                 inner = " ; ".join(checks)
                 argv = _ssh_base_argv(host, ssh_port) + [inner]
                 proc = await asyncio.create_subprocess_exec(
@@ -1562,9 +1499,6 @@ def setup_shell_routes() -> APIRouter:
                     in_container=_running_in_container() if not on_remote else False,
                     installed=pkg["installed"],
                     default_hint=pkg.get("install_hint"),
-                    host_docker_access=(
-                        _host_docker_access_enabled() if not on_remote else False
-                    ),
                 )
                 pkg["applicable"] = status.applicable
                 pkg["install_hint"] = status.install_hint
@@ -1600,7 +1534,6 @@ def setup_shell_routes() -> APIRouter:
             "onnxruntime",
             "hdbscan",
             "vllm",
-            "mlx-lm",
         }
         if pip_name not in known:
             return {"ok": False, "error": f"Unknown package: {pip_name}"}
@@ -1647,19 +1580,6 @@ def setup_shell_routes() -> APIRouter:
                 elif n == "g++": out += ["gcc-c++"]
                 else: out.append(n)
             return out
-        def _apk(names):
-            out = []
-            for n in names:
-                if n == "build-essential": out.append("build-base")
-                else: out.append(n)
-            return out
-        def _zypper(names):
-            out = []
-            for n in names:
-                if n == "build-essential": out += ["gcc-c++", "make"]
-                elif n == "g++": out.append("gcc-c++")
-                else: out.append(n)
-            return out
         def _brew(names):
             return [n for n in names if n not in ("build-essential", "g++", "gcc", "make")]
         # Build a single shell snippet that detects the package manager and
@@ -1668,8 +1588,6 @@ def setup_shell_routes() -> APIRouter:
         apt_pkgs = " ".join(shlex.quote(p) for p in _apt(pkgs))
         pac_pkgs = " ".join(shlex.quote(p) for p in _pacman(pkgs))
         dnf_pkgs = " ".join(shlex.quote(p) for p in _dnf(pkgs))
-        apk_pkgs = " ".join(shlex.quote(p) for p in _apk(pkgs))
-        zypper_pkgs = " ".join(shlex.quote(p) for p in _zypper(pkgs))
         brew_pkgs = " ".join(shlex.quote(p) for p in _brew(pkgs))
         # Error messages go to stderr (>&2) so the route's error field
         # gets populated. Without the redirect, `echo "ERROR…"` on stdout
@@ -1677,28 +1595,18 @@ def setup_shell_routes() -> APIRouter:
         # bare "HTTP 200" instead of surfacing the real reason.
         script = (
             'set -e; '
-            'BREW="$(command -v brew 2>/dev/null || true)"; '
-            'if [ -z "$BREW" ] && [ -x /opt/homebrew/bin/brew ]; then BREW=/opt/homebrew/bin/brew; fi; '
-            'if [ -z "$BREW" ] && [ -x /usr/local/bin/brew ]; then BREW=/usr/local/bin/brew; fi; '
-            'if [ -n "$BREW" ]; then '
-            f'  if [ -z "{brew_pkgs}" ]; then echo "Nothing to install with brew for requested packages." >&2; exit 4; fi; "$BREW" install {brew_pkgs}; exit $?; '
-            'fi; '
-            'if [ "$(id -u)" = "0" ]; then SUDO=""; '
-            'elif command -v sudo >/dev/null 2>&1 && sudo -n true 2>/dev/null; then SUDO="sudo -n"; '
-            'else '
-            '  echo "ERROR: this target needs sudo for its OS package manager, but passwordless sudo is unavailable. Open a terminal on the target and run the shown install command once, then retry in Cookbook." >&2; exit 2; fi; '
+            'if ! sudo -n true 2>/dev/null; then '
+            '  echo "ERROR: passwordless sudo unavailable on this target. Run once: sudo apt install -y ' + " ".join(pkgs) + ' (or your distro equivalent: pacman -S, dnf install, brew install). After that, Cookbook can install the rest." >&2; exit 2; fi; '
             'if command -v apt-get >/dev/null 2>&1; then '
-            f'  $SUDO env DEBIAN_FRONTEND=noninteractive apt-get update -qq && $SUDO env DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends {apt_pkgs}; '
+            f'  sudo -n env DEBIAN_FRONTEND=noninteractive apt-get update -qq && sudo -n env DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends {apt_pkgs}; '
             'elif command -v pacman >/dev/null 2>&1; then '
-            f'  $SUDO pacman -Sy --needed --noconfirm {pac_pkgs}; '
+            f'  sudo -n pacman -Sy --needed --noconfirm {pac_pkgs}; '
             'elif command -v dnf >/dev/null 2>&1; then '
-            f'  $SUDO dnf install -y {dnf_pkgs}; '
-            'elif command -v apk >/dev/null 2>&1; then '
-            f'  $SUDO apk add --no-interactive {apk_pkgs}; '
-            'elif command -v zypper >/dev/null 2>&1; then '
-            f'  $SUDO zypper --non-interactive install {zypper_pkgs}; '
+            f'  sudo -n dnf install -y {dnf_pkgs}; '
+            'elif command -v brew >/dev/null 2>&1; then '
+            f'  brew install {brew_pkgs}; '
             'else '
-            '  echo "ERROR: no supported package manager (apt/pacman/dnf/apk/zypper/brew) on this target." >&2; exit 3; fi'
+            '  echo "ERROR: no supported package manager (apt/pacman/dnf/brew) on this target." >&2; exit 3; fi'
         )
         try:
             if host:

routes/skills_routes.py

@@ -22,16 +22,6 @@ from core.middleware import require_admin
 
 logger = logging.getLogger(__name__)
 
-# Last-resort verdict extraction from a teacher/verifier model's prose (run when
-# JSON parsing fails). `["\'\s:]*` already consumes whitespace, so the original
-# trailing `\s*` made two adjacent \s-matching quantifiers that backtrack O(n^2)
-# on a `verdict` + whitespace flood in untrusted model output (CodeQL
-# py/polynomial-redos). Without it a single unbounded quantifier remains — the
-# matched text is identical, and the scan is linear.
-_VERDICT_PROSE_RE = re.compile(
-    r'verdict["\'\s:]*["\']?(pass|needs_work|fail|inconclusive)', re.I
-)
-
 
 class SkillAddRequest(BaseModel):
     # New schema (preferred)
@@ -206,7 +196,7 @@ async def _eval_skill_run(skill_md: str, task: str, transcript: str,
         # Last resort: pull the verdict keyword straight out of the prose so a
         # clearly-decided run isn't thrown away as "unparseable".
         if v not in _VERDICTS:
-            km = _VERDICT_PROSE_RE.search(text)
+            km = _re.search(r'verdict["\'\s:]*\s*["\']?(pass|needs_work|fail|inconclusive)', text, _re.I)
             if km:
                 v = km.group(1).lower()
                 if data is None:

routes/task_routes.py

@@ -11,14 +11,10 @@ from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel
 
 from core.database import SessionLocal, ScheduledTask, TaskRun
+from core.middleware import INTERNAL_TOOL_USER
 from core.constants import internal_api_base
 from src.auth_helpers import get_current_user
 from src.constants import DATA_DIR, EMAIL_URGENCY_CACHE_DIR
-from src.task_action_policy import (
-    ADMIN_ONLY_TASK_ACTIONS,
-    is_admin_only_task_action,
-    owner_has_admin_task_privileges,
-)
 from src.task_scheduler import compute_next_run, HOUSEKEEPING_DEFAULTS
 from routes.prefs_routes import _load_for_user, _save_for_user
 
@@ -421,18 +417,28 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 db.close()
         return {"ok": True, "opened": True, "enabled": bool(prefs.get("tasks_enabled")), "resumed": resumed}
 
-    # Actions that execute shell/SSH commands or cross into admin-only
-    # Cookbook serving surfaces — restricted to admins.
+    # Actions that execute shell/SSH commands — restricted to admins.
     # Non-admin users cannot create tasks with these action types via the
     # API. See review CRIT-C.
-    _ADMIN_ONLY_ACTIONS = ADMIN_ONLY_TASK_ACTIONS
+    _ADMIN_ONLY_ACTIONS = {"run_local", "run_script", "ssh_command"}
 
     def _is_admin(user: str | None) -> bool:
-        return owner_has_admin_task_privileges(user)
-
-    def _require_admin_for_task_action(user: str | None, task_type: str | None, action: str | None) -> None:
-        if is_admin_only_task_action(task_type, action) and not _is_admin(user):
-            raise HTTPException(403, f"Action '{action}' requires admin privileges")
+        if not user:
+            return False
+        # In-process tool-loopback marker — AuthMiddleware validated
+        # the internal token + loopback client before stamping this,
+        # so treat as admin-equivalent.
+        if user == INTERNAL_TOOL_USER:
+            return True
+        try:
+            from core.auth import AuthManager
+            auth = AuthManager()
+            if not auth.is_configured:
+                # Unconfigured single-user deploy: trust the local owner.
+                return True
+            return bool(auth.is_admin(user))
+        except Exception:
+            return False
 
     def _validate_then_task_id(db, then_task_id: Optional[str], user: Optional[str], current_task_id: Optional[str] = None) -> Optional[str]:
         target_id = (then_task_id or "").strip()
@@ -460,7 +466,8 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         # Block shell-executing action types for non-admins. action_run_local
         # uses subprocess.run(shell=True) and ssh_command / run_script run
         # arbitrary commands.
-        _require_admin_for_task_action(user, req.task_type, req.action)
+        if req.task_type == "action" and req.action in _ADMIN_ONLY_ACTIONS and not _is_admin(user):
+            raise HTTPException(403, f"Action '{req.action}' requires admin privileges")
         if req.trigger_type == "schedule" and not req.schedule:
             raise HTTPException(400, "Schedule is required for schedule-triggered tasks")
         if req.trigger_type == "schedule" and req.schedule == "cron" and not req.cron_expression:
@@ -587,7 +594,6 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         cache_tables = {
             "summarize_emails": ("email_summaries",),
             "draft_email_replies": ("email_ai_replies",),
-            "email_auto_translate": ("email_translations",),
             "extract_email_events": ("email_calendar_extractions",),
             "learn_sender_signatures": ("sender_signatures",),
             "check_email_urgency": ("email_tags", "email_urgency_alerts"),
@@ -674,10 +680,6 @@ def setup_task_routes(task_scheduler) -> APIRouter:
             if user and task.owner != user:
                 raise HTTPException(403, "Access denied")
 
-            next_task_type = req.task_type if req.task_type is not None else task.task_type
-            next_action = req.action if req.action is not None else task.action
-            _require_admin_for_task_action(user, next_task_type, next_action)
-
             if req.name is not None:
                 task.name = req.name
             if req.prompt is not None:
@@ -685,6 +687,9 @@ def setup_task_routes(task_scheduler) -> APIRouter:
             if req.task_type is not None:
                 task.task_type = req.task_type
             if req.action is not None:
+                # Same admin-only gate as create — see CRIT-C.
+                if req.action in _ADMIN_ONLY_ACTIONS and not _is_admin(user):
+                    raise HTTPException(403, f"Action '{req.action}' requires admin privileges")
                 task.action = req.action
             if req.output_target is not None:
                 task.output_target = req.output_target
@@ -801,7 +806,6 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 raise HTTPException(404, "Task not found")
             if user and task.owner != user:
                 raise HTTPException(403, "Access denied")
-            _require_admin_for_task_action(user, task.task_type, task.action)
             task.status = "active"
             if (task.trigger_type or "schedule") == "schedule":
                 task.next_run = compute_next_run(
@@ -864,7 +868,6 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 raise HTTPException(404, "Task not found")
             if user and task.owner != user:
                 raise HTTPException(403, "Access denied")
-            _require_admin_for_task_action(user, task.task_type, task.action)
         finally:
             db.close()
         started = await task_scheduler.run_task_now(task_id, force=force)
@@ -890,11 +893,10 @@ def setup_task_routes(task_scheduler) -> APIRouter:
         return {"ok": True, "message": "Task stopped"}
 
     @router.get("/runs/recent")
-    async def list_recent_runs(request: Request, limit: int = 50, max_result_chars: int = 6000):
+    async def list_recent_runs(request: Request, limit: int = 50):
         """Recent task runs across ALL tasks for this owner. Drives the Activity view."""
         user = _owner(request)
         limit = max(1, min(limit, 200))
-        max_result_chars = max(500, min(max_result_chars, 20000))
         db = SessionLocal()
         try:
             q = db.query(TaskRun, ScheduledTask).join(
@@ -928,20 +930,10 @@ def setup_task_routes(task_scheduler) -> APIRouter:
                 deduped.append((r, t))
                 if len(deduped) >= limit:
                     break
-
-            def _clip_run(r: TaskRun) -> dict:
-                d = _run_to_dict(r)
-                for key in ("result", "error"):
-                    val = d.get(key)
-                    if isinstance(val, str) and len(val) > max_result_chars:
-                        d[key] = val[:max_result_chars].rstrip() + "\n\n[Activity preview truncated]"
-                return d
-
             return {
-                "has_more": len(rows) > len(deduped),
                 "runs": [
                     {
-                        **_clip_run(r),
+                        **_run_to_dict(r),
                         "task_name": _display_task_name(t),
                         "task_type": t.task_type or "llm",
                         "action": t.action,
@@ -1054,14 +1046,6 @@ def setup_task_routes(task_scheduler) -> APIRouter:
             ).first()
             if not task:
                 raise HTTPException(404, "Not found")
-            if (
-                is_admin_only_task_action(task.task_type, task.action)
-                and not owner_has_admin_task_privileges(task.owner)
-            ):
-                task.status = "paused"
-                task.next_run = None
-                db.commit()
-                raise HTTPException(403, f"Action '{task.action}' requires admin privileges")
         finally:
             db.close()
         started = await task_scheduler.run_task_now(task_id)

routes/upload_routes.py

@@ -6,11 +6,11 @@ import asyncio
 import shutil
 import uuid
 from pathlib import Path
-from fastapi import APIRouter, Request, File, UploadFile, HTTPException, Form
-from typing import List, Optional
+from fastapi import APIRouter, Request, File, UploadFile, HTTPException
+from typing import List
 import logging
 from core.middleware import require_admin
-from core.database import SessionLocal, GalleryImage, Session as DbSession
+from core.database import SessionLocal, GalleryImage
 from src.auth_helpers import effective_user
 from src.constants import GENERATED_IMAGES_DIR
 from src.upload_handler import count_recent_uploads
@@ -56,17 +56,7 @@ def setup_upload_routes(upload_handler):
 
         raise HTTPException(404, "File not found")
 
-    def _valid_session_id_for_owner(db, session_id: str | None, owner: str | None) -> str | None:
-        if not session_id:
-            return None
-        sess = db.query(DbSession).filter(DbSession.id == session_id).first()
-        if not sess:
-            return None
-        if owner and sess.owner and sess.owner != owner:
-            return None
-        return session_id
-
-    def _promote_chat_image_to_gallery(meta: dict, owner: str | None, session_id: str | None = None) -> str | None:
+    def _promote_chat_image_to_gallery(meta: dict, owner: str | None) -> str | None:
         """Make chat-uploaded images visible in Gallery without changing chat storage."""
         is_image_file = getattr(upload_handler, "is_image_file", None)
         if not callable(is_image_file):
@@ -115,7 +105,6 @@ def setup_upload_routes(upload_handler):
                 prompt=meta.get("name") or "Chat upload",
                 model="chat-upload",
                 owner=owner,
-                session_id=_valid_session_id_for_owner(db, session_id, owner),
                 file_hash=file_hash,
                 width=meta.get("width"),
                 height=meta.get("height"),
@@ -131,14 +120,8 @@ def setup_upload_routes(upload_handler):
             db.close()
     
     @router.post("")
-    async def api_upload(
-        request: Request,
-        files: List[UploadFile] = File(...),
-        session_id: Optional[str] = Form(None),
-    ):
+    async def api_upload(request: Request, files: List[UploadFile] = File(...)):
         """Upload files with enhanced security and organization."""
-        if not isinstance(session_id, str):
-            session_id = None
         if not files:
             raise HTTPException(400, "No files uploaded")
             
@@ -165,7 +148,7 @@ def setup_upload_routes(upload_handler):
             try:
                 owner = effective_user(request)
                 meta = upload_handler.save_upload(u, client_ip, owner=owner)
-                gallery_id = _promote_chat_image_to_gallery(meta, owner, session_id)
+                gallery_id = _promote_chat_image_to_gallery(meta, owner)
                 item = {
                     "id": meta["id"],
                     "name": meta["name"],
@@ -218,13 +201,14 @@ def setup_upload_routes(upload_handler):
         import mimetypes as _mt
         # Look up original filename and owner from uploads.json
         original_name = file_id
-        # _load_upload_index() tolerates a missing/corrupt uploads.json (it falls
-        # back to the .bak sibling, then to {}), so a truncated DB degrades to
-        # "no metadata" instead of a 500 from an unhandled JSONDecodeError.
-        db = upload_handler._load_upload_index()
-        info = next((fi for fi in db.values() if fi.get("id") == file_id), None)
-        if info:
-            original_name = info.get("name", file_id)
+        info = None
+        uploads_db = os.path.join(_upload_root(), "uploads.json")
+        if os.path.exists(uploads_db):
+            with open(uploads_db, encoding="utf-8") as f:
+                db = json.load(f)
+            info = next((fi for fi in db.values() if fi.get("id") == file_id), None)
+            if info:
+                original_name = info.get("name", file_id)
         auth_mgr = getattr(request.app.state, "auth_manager", None)
         auth_configured = bool(auth_mgr and auth_mgr.is_configured)
         current_user = effective_user(request)
@@ -270,42 +254,19 @@ def setup_upload_routes(upload_handler):
 
     def _load_upload_info(file_id: str):
         """Look up the uploads.json record for a file_id, with owner/auth checks."""
-        # Corruption-tolerant load (see download_file): a bad uploads.json yields
-        # {} rather than raising JSONDecodeError out of the vision path.
-        db = upload_handler._load_upload_index()
-        return next((fi for fi in db.values() if fi.get("id") == file_id), None)
+        info = None
+        uploads_db = os.path.join(_upload_root(), "uploads.json")
+        if os.path.exists(uploads_db):
+            with open(uploads_db, encoding="utf-8") as f:
+                db = json.load(f)
+            info = next((fi for fi in db.values() if fi.get("id") == file_id), None)
+        return info
 
     def _vision_cache_path(file_id: str) -> str:
         cache_dir = os.path.join(_upload_root(), ".vision")
         os.makedirs(cache_dir, exist_ok=True)
         return os.path.join(cache_dir, file_id + ".txt")
 
-    def _sync_gallery_caption_for_upload(info: dict | None, owner: str | None, text: str) -> None:
-        """Copy upload OCR/vision text onto the promoted gallery image row."""
-        if not info:
-            return
-        file_hash = info.get("hash")
-        if not file_hash:
-            return
-        db = SessionLocal()
-        try:
-            q = db.query(GalleryImage).filter(
-                GalleryImage.file_hash == file_hash,
-                GalleryImage.is_active == True,  # noqa: E712
-            )
-            if owner:
-                q = q.filter(GalleryImage.owner == owner)
-            img = q.first()
-            if not img:
-                return
-            img.caption = (text or "").strip()
-            db.commit()
-        except Exception as e:
-            db.rollback()
-            logger.warning("Failed to sync OCR caption to gallery image: %s", e)
-        finally:
-            db.close()
-
     @router.get("/{file_id}/vision")
     async def get_vision_text(request: Request, file_id: str, force: int = 0):
         """Return the vision-model OCR/description for an uploaded image.
@@ -332,9 +293,7 @@ def setup_upload_routes(upload_handler):
         if not force and os.path.exists(cache_path):
             try:
                 with open(cache_path, encoding="utf-8") as f:
-                    cached_text = f.read()
-                _sync_gallery_caption_for_upload(info, file_owner or current_user, cached_text)
-                return {"text": cached_text, "cached": True}
+                    return {"text": f.read(), "cached": True}
             except Exception as e:
                 logger.warning(f"Vision cache read failed for {file_id}: {e}")
         from src.document_processor import analyze_image_with_vl
@@ -348,7 +307,6 @@ def setup_upload_routes(upload_handler):
                 f.write(text)
         except Exception as e:
             logger.warning(f"Vision cache write failed for {file_id}: {e}")
-        _sync_gallery_caption_for_upload(info, file_owner or current_user, text)
         return {"text": text, "cached": False}
 
     @router.put("/{file_id}/vision")
@@ -370,16 +328,12 @@ def setup_upload_routes(upload_handler):
             if file_owner != current_user and not auth_mgr.is_admin(current_user):
                 raise HTTPException(404, "File not found")
         _resolve_upload_path(file_id)
-        try:
-            body = await request.json()
-        except json.JSONDecodeError:
-            raise HTTPException(400, "Request body must be valid JSON")
+        body = await request.json()
         text = (body or {}).get("text", "")
         if not isinstance(text, str):
             raise HTTPException(400, "text must be a string")
         with open(_vision_cache_path(file_id), "w", encoding="utf-8") as f:
             f.write(text)
-        _sync_gallery_caption_for_upload(info, file_owner or current_user, text)
         return {"ok": True}
 
     async def periodic_rate_limit_cleanup():

routes/webhook_routes.py

@@ -345,9 +345,8 @@ def setup_webhook_routes(
                             resp = await client.get(models_url, headers=hdrs)
                             resp.raise_for_status()
                             data = resp.json()
-                            items = data if isinstance(data, list) else (data.get("data") or [])
-                            ids = [m.get("id") for m in items if isinstance(m, dict) and m.get("id")]
-                            if not ids and isinstance(data, dict):
+                            ids = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
+                            if not ids:
                                 ids = [
                                     m.get("name") or m.get("model")
                                     for m in (data.get("models") or [])

scripts/claim_ownerless.py

@@ -27,18 +27,12 @@ def claim_json_entries(entries, owner):
     return count
 
 
-def owner_arg(argv):
-    if len(argv) < 2 or not argv[1].strip():
-        return None
-    return argv[1].strip()
-
-
 def main():
-    owner = owner_arg(sys.argv)
-    if not owner:
+    if len(sys.argv) < 2:
         print("Usage: python scripts/claim_ownerless.py <username>")
         sys.exit(1)
 
+    owner = sys.argv[1]
     print(f"Claiming all ownerless data for: {owner}\n")
 
     # 1. Memories (JSON files)

scripts/odysseus-gallery

@@ -38,27 +38,23 @@ def _preview_text(value, limit: int = 200) -> str:
     return text[:limit]
 
 
-def _text_field(value) -> str:
-    return value if isinstance(value, str) else ""
-
-
 def _serialize_image(i: "GalleryImage") -> dict:
     return {
         "id": i.id,
-        "filename": _text_field(i.filename),
+        "filename": i.filename,
         "prompt": _preview_text(i.prompt),
-        "model": _text_field(i.model),
-        "size": _text_field(i.size),
-        "tags": _text_field(i.tags),
+        "model": i.model or "",
+        "size": i.size or "",
+        "tags": i.tags or "",
         "favorite": bool(i.favorite),
-        "album_id": _text_field(i.album_id),
-        "session_id": _text_field(i.session_id),
+        "album_id": i.album_id or "",
+        "session_id": i.session_id or "",
         "width": i.width,
         "height": i.height,
         "file_size": i.file_size,
         "taken_at": i.taken_at.isoformat() if i.taken_at else "",
-        "camera_make": _text_field(i.camera_make),
-        "camera_model": _text_field(i.camera_model),
+        "camera_make": i.camera_make or "",
+        "camera_model": i.camera_model or "",
         "created_at": i.created_at.isoformat() if i.created_at else "",
     }
 
@@ -97,11 +93,11 @@ def cmd_show(args):
         if not i:
             fail(f"no image with id {args.id!r}")
         out = _serialize_image(i)
-        out["prompt_full"] = _text_field(i.prompt)
-        out["ai_tags"] = _text_field(i.ai_tags)
+        out["prompt_full"] = i.prompt or ""
+        out["ai_tags"] = i.ai_tags or ""
         out["gps_lat"] = i.gps_lat or ""
         out["gps_lng"] = i.gps_lng or ""
-        out["file_hash"] = _text_field(i.file_hash)
+        out["file_hash"] = i.file_hash or ""
         emit(out, args)
     finally:
         db.close()

scripts/odysseus-mail

@@ -108,8 +108,6 @@ def _q(name: str) -> str:
 
 
 def _split_recipients(value: str) -> list[str]:
-    if not isinstance(value, str):
-        return []
     return [r.strip() for r in (value or "").split(",") if r.strip()]
 
 

scripts/odysseus-notes

@@ -36,9 +36,7 @@ def _load_items(raw) -> list:
         items = json.loads(raw)
     except (TypeError, json.JSONDecodeError):
         return []
-    if not isinstance(items, list):
-        return []
-    return [item for item in items if isinstance(item, dict)]
+    return items if isinstance(items, list) else []
 
 
 def _serialize(n: "Note") -> dict:

services/hwfit/data/hf_models.json

@@ -5113,8 +5113,8 @@
  {
   "name": "deepseek-ai/DeepSeek-V4-Flash",
   "provider": "deepseek-ai",
-  "parameter_count": "158.1B",
-  "parameters_raw": 158069433298,
+  "parameter_count": "284B",
+  "parameters_raw": 284000000000,
   "active_parameters": 13000000000,
   "is_moe": true,
   "min_ram_gb": 200.0,
@@ -5130,40 +5130,15 @@
   ],
   "pipeline_tag": "text-generation",
   "architecture": "deepseek_v4_moe",
-  "hf_downloads": 1882337,
-  "hf_likes": 1651,
-  "release_date": "2026-06-22"
- },
- {
-  "name": "deepseek-ai/DeepSeek-V4-Flash-DSpark",
-  "provider": "deepseek-ai",
-  "parameter_count": "165.3B",
-  "parameters_raw": 165265454782,
-  "active_parameters": 13000000000,
-  "is_moe": true,
-  "active_experts": 6,
-  "min_ram_gb": 170.0,
-  "recommended_ram_gb": 250.0,
-  "min_vram_gb": 165.0,
-  "quantization": "FP8-Mixed",
-  "context_length": 1000000,
-  "use_case": "General-purpose reasoning, long-context",
-  "capabilities": [
-   "long_context",
-   "reasoning",
-   "moe"
-  ],
-  "pipeline_tag": "text-generation",
-  "architecture": "deepseek_v4_moe",
-  "hf_downloads": 4446,
-  "hf_likes": 107,
-  "release_date": "2026-06-27"
+  "hf_downloads": 3542202,
+  "hf_likes": 0,
+  "release_date": "2026-05-15"
  },
  {
   "name": "deepseek-ai/DeepSeek-V4-Flash-Base",
   "provider": "deepseek-ai",
-  "parameter_count": "292.0B",
-  "parameters_raw": 292021347282,
+  "parameter_count": "284B",
+  "parameters_raw": 284000000000,
   "active_parameters": 13000000000,
   "is_moe": true,
   "min_ram_gb": 290.0,
@@ -5178,15 +5153,15 @@
   ],
   "pipeline_tag": "text-generation",
   "architecture": "deepseek_v4_moe",
-  "hf_downloads": 76030,
-  "hf_likes": 256,
-  "release_date": "2026-04-27"
+  "hf_downloads": 0,
+  "hf_likes": 0,
+  "release_date": "2026-05-15"
  },
  {
   "name": "deepseek-ai/DeepSeek-V4-Pro",
   "provider": "deepseek-ai",
-  "parameter_count": "861.6B",
-  "parameters_raw": 861608274846,
+  "parameter_count": "1.6T",
+  "parameters_raw": 1600000000000,
   "active_parameters": 49000000000,
   "is_moe": true,
   "min_ram_gb": 1100.0,
@@ -5202,40 +5177,15 @@
   ],
   "pipeline_tag": "text-generation",
   "architecture": "deepseek_v4_moe",
-  "hf_downloads": 1154610,
-  "hf_likes": 5118,
-  "release_date": "2026-06-22"
- },
- {
-  "name": "deepseek-ai/DeepSeek-V4-Pro-DSpark",
-  "provider": "deepseek-ai",
-  "parameter_count": "889.5B",
-  "parameters_raw": 889484881098,
-  "active_parameters": 49000000000,
-  "is_moe": true,
-  "active_experts": 6,
-  "min_ram_gb": 900.0,
-  "recommended_ram_gb": 1250.0,
-  "min_vram_gb": 890.0,
-  "quantization": "FP8-Mixed",
-  "context_length": 1000000,
-  "use_case": "Flagship reasoning, long-context",
-  "capabilities": [
-   "long_context",
-   "reasoning",
-   "moe"
-  ],
-  "pipeline_tag": "text-generation",
-  "architecture": "deepseek_v4_moe",
-  "hf_downloads": 6939,
-  "hf_likes": 241,
-  "release_date": "2026-06-27"
+  "hf_downloads": 0,
+  "hf_likes": 0,
+  "release_date": "2026-05-15"
  },
  {
   "name": "deepseek-ai/DeepSeek-V4-Pro-Base",
   "provider": "deepseek-ai",
   "parameter_count": "1.6T",
-  "parameters_raw": 1600790440862,
+  "parameters_raw": 1600000000000,
   "active_parameters": 49000000000,
   "is_moe": true,
   "min_ram_gb": 1700.0,
@@ -5250,9 +5200,9 @@
   ],
   "pipeline_tag": "text-generation",
   "architecture": "deepseek_v4_moe",
-  "hf_downloads": 25387,
-  "hf_likes": 305,
-  "release_date": "2026-04-27"
+  "hf_downloads": 0,
+  "hf_likes": 0,
+  "release_date": "2026-05-15"
  },
  {
   "name": "deepseek-ai/deepseek-coder-6.7b-base",
@@ -13358,106 +13308,6 @@
   "_discovered": true,
   "gguf_sources": []
  },
- {
-  "name": "zai-org/GLM-5.2",
-  "provider": "zai-org",
-  "parameter_count": "753.3B",
-  "parameters_raw": 753329940480,
-  "min_ram_gb": 1510.0,
-  "recommended_ram_gb": 1800.0,
-  "min_vram_gb": 1510.0,
-  "quantization": "BF16",
-  "context_length": 1048576,
-  "use_case": "General purpose reasoning, coding, long-context",
-  "capabilities": [
-   "long_context",
-   "reasoning",
-   "coding",
-   "moe"
-  ],
-  "pipeline_tag": "text-generation",
-  "architecture": "glm_moe_dsa",
-  "hf_downloads": 142547,
-  "hf_likes": 2996,
-  "release_date": "2026-06-23",
-  "is_moe": true,
-  "active_experts": 8,
-  "gguf_sources": [
-   {
-    "repo": "unsloth/GLM-5.2-GGUF",
-    "provider": "unsloth",
-    "file": "UD-Q4_K_M/*.gguf",
-    "quant": "Q4_K_M"
-   }
-  ]
- },
- {
-  "name": "zai-org/GLM-5.2-FP8",
-  "provider": "zai-org",
-  "parameter_count": "753.4B",
-  "parameters_raw": 753375793584,
-  "min_ram_gb": 760.0,
-  "recommended_ram_gb": 900.0,
-  "min_vram_gb": 760.0,
-  "quantization": "FP8",
-  "context_length": 1048576,
-  "use_case": "General purpose reasoning, coding, long-context",
-  "capabilities": [
-   "long_context",
-   "reasoning",
-   "coding",
-   "moe"
-  ],
-  "pipeline_tag": "text-generation",
-  "architecture": "glm_moe_dsa",
-  "hf_downloads": 884226,
-  "hf_likes": 182,
-  "release_date": "2026-06-23",
-  "is_moe": true,
-  "active_experts": 8,
-  "gguf_sources": [
-   {
-    "repo": "unsloth/GLM-5.2-GGUF",
-    "provider": "unsloth",
-    "file": "UD-Q4_K_M/*.gguf",
-    "quant": "Q4_K_M"
-   }
-  ]
- },
- {
-  "name": "unsloth/GLM-5.2-GGUF",
-  "provider": "unsloth",
-  "parameter_count": "753.9B",
-  "parameters_raw": 753864139008,
-  "min_ram_gb": 452.0,
-  "recommended_ram_gb": 620.0,
-  "min_vram_gb": 452.0,
-  "quantization": "Q4_K_M",
-  "context_length": 1048576,
-  "use_case": "General purpose reasoning, coding, long-context (GGUF)",
-  "capabilities": [
-   "long_context",
-   "reasoning",
-   "coding",
-   "moe"
-  ],
-  "pipeline_tag": "text-generation",
-  "architecture": "glm-dsa",
-  "hf_downloads": 180394,
-  "hf_likes": 474,
-  "release_date": "2026-06-23",
-  "is_moe": true,
-  "active_experts": 8,
-  "is_gguf": true,
-  "gguf_sources": [
-   {
-    "repo": "unsloth/GLM-5.2-GGUF",
-    "provider": "unsloth",
-    "file": "UD-Q4_K_M/*.gguf",
-    "quant": "Q4_K_M"
-   }
-  ]
- },
  {
   "name": "cyankiwi/Qwen3.5-35B-A3B-AWQ-4bit",
   "provider": "cyankiwi",
@@ -19105,54 +18955,6 @@
   "active_experts": 8,
   "active_parameters": 13600000000
  },
- {
-  "name": "MiniMaxAI/MiniMax-M3",
-  "provider": "MiniMaxAI",
-  "parameter_count": "427.0B",
-  "parameters_raw": 427040140160,
-  "min_ram_gb": 855.0,
-  "recommended_ram_gb": 1025.0,
-  "min_vram_gb": 855.0,
-  "quantization": "BF16",
-  "context_length": 1000000,
-  "use_case": "Vision, chat, coding, agentic tool use",
-  "capabilities": [
-   "vision",
-   "tool_use",
-   "coding",
-   "moe"
-  ],
-  "pipeline_tag": "image-text-to-text",
-  "architecture": "minimax_m3_vl",
-  "hf_downloads": 192311,
-  "hf_likes": 1267,
-  "release_date": "2026-06-23",
-  "is_moe": true
- },
- {
-  "name": "MiniMaxAI/MiniMax-M3-MXFP8",
-  "provider": "MiniMaxAI",
-  "parameter_count": "440.3B",
-  "parameters_raw": 440279845760,
-  "min_ram_gb": 445.0,
-  "recommended_ram_gb": 560.0,
-  "min_vram_gb": 445.0,
-  "quantization": "MXFP8",
-  "context_length": 1000000,
-  "use_case": "Vision, chat, coding, agentic tool use",
-  "capabilities": [
-   "vision",
-   "tool_use",
-   "coding",
-   "moe"
-  ],
-  "pipeline_tag": "image-text-to-text",
-  "architecture": "minimax_m3_vl",
-  "hf_downloads": 572278,
-  "hf_likes": 43,
-  "release_date": "2026-06-15",
-  "is_moe": true
- },
  {
   "name": "bullerwins/MiniMax-M2.7-REAP-172B-fp8",
   "provider": "bullerwins",
@@ -19474,4 +19276,4 @@
   ],
   "_discovered": true
  }
-]
+]

services/hwfit/fit.py

@@ -168,19 +168,6 @@ def _canonical_cpu_backend(system):
     return "cpu_x86"
 
 
-def _is_mlx_model(model, native_q=None):
-    name = (model.get("name") or "").lower()
-    provider = (model.get("provider") or "").lower()
-    fmt = (model.get("format") or "").lower()
-    q = (native_q if native_q is not None else _native_quant(model)).lower()
-    return (
-        q.startswith("mlx-")
-        or provider == "mlx-community"
-        or fmt == "mlx"
-        or name.startswith("mlx-community/")
-    )
-
-
 def _estimate_speed(model, quant, run_mode, system, offload_frac=0.0):
     """Estimate tok/s. Uses active params for MoE (only active experts run per token).
 
@@ -326,22 +313,6 @@ def _fit_score(required, available):
     return 50
 
 
-def _is_unified_memory_system(system):
-    backend = (system.get("backend") or "").lower()
-    return bool(system.get("unified_memory")) or backend in ("metal", "mps", "apple")
-
-
-def _fit_level_for_budget(required_gb, budget_gb):
-    if not required_gb or not budget_gb or required_gb > budget_gb:
-        return "too_tight"
-    ratio = required_gb / budget_gb
-    if ratio <= 0.50:
-        return "perfect"
-    if ratio <= 0.78:
-        return "good"
-    return "marginal"
-
-
 def _context_score(ctx, use_case):
     target = CONTEXT_TARGET.get(use_case, 4096)
     if ctx >= target:
@@ -545,42 +516,21 @@ def analyze_model(model, system, target_quant=None, scoring_use_case=None, targe
     run_mode, quant, fit_ctx, required_gb = result
 
     # Determine fit level
-    unified_memory = _is_unified_memory_system(system)
-    total_ram = system.get("total_ram_gb") or available_ram
-    unified_budget = max(total_ram or 0, available_ram or 0, effective_vram or 0)
-    budget = unified_budget if unified_memory else (effective_vram if run_mode == "gpu" else available_ram)
+    budget = effective_vram if run_mode == "gpu" else available_ram
     if required_gb > budget:
         return None
     if run_mode == "gpu":
-        if unified_memory:
-            fit_level = _fit_level_for_budget(required_gb, budget)
-        else:
-            # GPU-only fit must leave real allocator/KV/runtime headroom. The
-            # old check used recommended_ram_gb (or required_gb as a fallback),
-            # which made any model that barely fit VRAM read as "perfect".
-            # On CUDA/vLLM/SGLang that is misleading: 141 GB on a 160 GB box is
-            # runnable, but not a comfortable perfect fit.
-            if gpu_vram >= required_gb * 1.50:
-                fit_level = "perfect"
-            elif gpu_vram >= required_gb * 1.2:
-                fit_level = "good"
-            else:
-                fit_level = "marginal"
-    elif run_mode == "cpu_offload":
-        fit_level = _fit_level_for_budget(required_gb, budget)
-        if fit_level == "perfect":
+        rec = model.get("recommended_ram_gb") or required_gb
+        if rec <= gpu_vram:
+            fit_level = "perfect"
+        elif gpu_vram >= required_gb * 1.2:
             fit_level = "good"
-    else:
-        fit_level = _fit_level_for_budget(required_gb, budget)
-        if fit_level == "too_tight":
+        else:
             fit_level = "marginal"
-
-    # Rows that comfortably fit in a huge RAM/unified-memory pool should not all
-    # look "marginal"; that made 1B-70B CPU/Ollama rows orange on 256 GB systems.
-    if fit_level == "marginal" and budget and required_gb <= budget * 0.78:
-        fit_level = "good"
-    if fit_level == "good" and budget and required_gb <= budget * 0.50 and run_mode != "cpu_offload":
-        fit_level = "perfect"
+    elif run_mode == "cpu_offload":
+        fit_level = "good" if available_ram >= required_gb * 1.2 else "marginal"
+    else:
+        fit_level = "marginal"
 
     # Fraction of the model that spills to CPU RAM (drives the offload speed
     # model). When offloading, anything beyond the GPU's VRAM lives in system RAM.
@@ -671,40 +621,6 @@ SORT_KEYS = {
 }
 
 
-def _search_blob(*parts):
-    text = " ".join(str(p or "") for p in parts).lower()
-    compact = re.sub(r"[^a-z0-9]+", "", text)
-    spaced = re.sub(r"[^a-z0-9]+", " ", text).strip()
-    return f"{text} {spaced} {compact}"
-
-
-def _matches_search(model, search):
-    terms = [t for t in re.split(r"\s+", (search or "").strip().lower()) if t]
-    if not terms:
-        return True
-    blob = _search_blob(
-        model.get("name"),
-        model.get("provider"),
-        model.get("architecture"),
-        model.get("quantization"),
-        model.get("format"),
-        model.get("parameter_count"),
-    )
-    for term in terms:
-        norm = re.sub(r"[^a-z0-9]+", "", term)
-        if term not in blob and (not norm or norm not in blob):
-            if re.fullmatch(r"\d+(?:\.\d+)?b?", term):
-                try:
-                    wanted = float(term.rstrip("b"))
-                    actual = params_b(model)
-                except (TypeError, ValueError):
-                    actual = 0
-                if wanted > 0 and actual > 0 and abs(actual - wanted) <= max(5.0, wanted * 0.08):
-                    continue
-            return False
-    return True
-
-
 def rank_models(system, use_case=None, limit=50, search=None, sort="score", quant=None, target_context=None, fit_only=False):
     """Rank all models against detected hardware. Returns sorted list of fit results.
 
@@ -777,11 +693,10 @@ def rank_models(system, use_case=None, limit=50, search=None, sort="score", quan
 
     for m in models:
         native_q = _native_quant(m)
-        is_mlx = _is_mlx_model(m, native_q)
 
-        # MLX is Apple Silicon-only. It should never appear on CUDA/ROCm/CPU,
-        # but it is first-class on Metal where mlx_lm.server can serve it.
-        if is_mlx and not apple_silicon:
+        # MLX needs the mlx_lm runtime, which Odysseus does not generate serve
+        # commands for. Hide it on every backend, including Metal.
+        if native_q.startswith("mlx-") or "mlx" in (m.get("name") or "").lower():
             continue
 
         # ROCm support for vLLM/SGLang quantized safetensors is too brittle to
@@ -808,7 +723,7 @@ def rank_models(system, use_case=None, limit=50, search=None, sort="score", quan
         # Windows is the same: Odysseus only supports llama.cpp on Windows,
         # which requires GGUF. vLLM/SGLang are explicitly blocked, so AWQ/GPTQ
         # models without a GGUF source are unservable there.
-        if (apple_silicon or consumer_amd or is_windows) and not is_mlx and not (m.get("is_gguf") or m.get("gguf_sources")):
+        if (apple_silicon or consumer_amd or is_windows) and not (m.get("is_gguf") or m.get("gguf_sources")):
             continue
 
         # Format filter: AWQ tab -> only AWQ models, FP4 tab -> FP4-family models, etc.
@@ -826,26 +741,13 @@ def rank_models(system, use_case=None, limit=50, search=None, sort="score", quan
             if quant in ("INT4", "INT8", "W4A16", "W8A8", "W8A16") and native_q != quant:
                 continue
 
-        if search and not _matches_search(m, search):
-            continue
+        if search:
+            name = m.get("name", "").lower()
+            provider = m.get("provider", "").lower()
+            if search.lower() not in name and search.lower() not in provider:
+                continue
 
-        model_quant = quant
-        # UI "Q4" means the user's looking for a 4-bit fit. On multi-GPU
-        # CUDA/vLLM/SGLang boxes, many practical 4-bit models are native AWQ
-        # safetensors, not GGUF Q4_K_M. If we pass Q4_K_M into a prequantized
-        # AWQ row, analyze_model correctly rejects it as the wrong serving
-        # format, but the result is confusing: highlighting Quant/Q4 hides the
-        # exact AWQ rows the machine is built to run. Treat Q4 as AWQ-4bit for
-        # native AWQ rows only on accelerator servers that can serve them.
-        if (
-            quant == "Q4_K_M"
-            and system.get("gpu_count", 1) >= 2
-            and not (apple_silicon or consumer_amd or is_windows)
-            and native_q == "AWQ-4bit"
-        ):
-            model_quant = native_q
-
-        result = analyze_model(m, system, target_quant=model_quant, scoring_use_case=(use_case or "general"), target_context=target_context)
+        result = analyze_model(m, system, target_quant=quant, scoring_use_case=(use_case or "general"), target_context=target_context)
         if result is None:
             continue
 

services/hwfit/hf_discovery.py

@@ -1,374 +0,0 @@
-import json
-import os
-import re
-import time
-import urllib.parse
-import urllib.request
-from email.utils import parsedate_to_datetime
-from pathlib import Path
-
-from src.constants import DATA_DIR
-
-
-HF_COLLECTIONS_URL = "https://huggingface.co/api/collections"
-HW_FIT_CACHE_DIR = Path(DATA_DIR) / "hwfit"
-MLX_COMMUNITY_CACHE = HW_FIT_CACHE_DIR / "mlx_community_models.json"
-HF_COLLECTION_MODELS_CACHE = HW_FIT_CACHE_DIR / "hf_collection_models.json"
-HF_COLLECTION_TTL_SECONDS = 24 * 3600
-
-
-HF_COLLECTION_SOURCES = (
-    {
-        "key": "mlx_community",
-        "owner": "mlx-community",
-        "provider": "mlx-community",
-        "repo_prefix": "mlx-community/",
-        "mlx_only": True,
-    },
-    {
-        "key": "zai_org",
-        "owner": "zai-org",
-        "provider": "zai-org",
-    },
-    {
-        "key": "deepseek_ai",
-        "owner": "deepseek-ai",
-        "provider": "deepseek-ai",
-    },
-    {
-        "key": "minimax_ai",
-        "owner": "MiniMaxAI",
-        "provider": "MiniMaxAI",
-    },
-    {
-        "key": "qwen",
-        "owner": "Qwen",
-        "provider": "Qwen",
-    },
-    {
-        "key": "stepfun_ai",
-        "owner": "stepfun-ai",
-        "provider": "stepfun-ai",
-    },
-    {
-        "key": "google",
-        "owner": "google",
-        "provider": "google",
-    },
-    {
-        "key": "openai",
-        "owner": "openai",
-        "provider": "openai",
-    },
-    {
-        "key": "mistralai",
-        "owner": "mistralai",
-        "provider": "mistralai",
-    },
-    {
-        "key": "meta_llama",
-        "owner": "meta-llama",
-        "provider": "meta-llama",
-    },
-    {
-        "key": "nousresearch",
-        "owner": "NousResearch",
-        "provider": "NousResearch",
-    },
-    {
-        "key": "moonshotai",
-        "owner": "moonshotai",
-        "provider": "moonshotai",
-    },
-    {
-        "key": "mllama",
-        "owner": "mllama",
-        "provider": "mllama",
-    },
-)
-
-
-def _format_params(raw):
-    try:
-        n = int(raw or 0)
-    except (TypeError, ValueError):
-        n = 0
-    if n <= 0:
-        return "", 0
-    if n >= 1_000_000_000_000:
-        return f"{n / 1_000_000_000_000:.3g}T", n
-    if n >= 1_000_000_000:
-        return f"{n / 1_000_000_000:.4g}B", n
-    if n >= 1_000_000:
-        return f"{n / 1_000_000:.4g}M", n
-    if n >= 1_000:
-        return f"{n / 1_000:.4g}K", n
-    return str(n), n
-
-
-def _parse_params_from_name(repo_id):
-    name = (repo_id or "").rsplit("/", 1)[-1]
-    active = None
-    m_active = re.search(r"[-_][Aa](\d+(?:\.\d+)?)[Bb](?![a-zA-Z])", name)
-    if m_active:
-        active = int(float(m_active.group(1)) * 1_000_000_000)
-        name = name[: m_active.start()] + name[m_active.end() :]
-    total = None
-    for m in re.finditer(r"(\d+(?:\.\d+)?)[Bb](?![a-zA-Z])", name):
-        total = int(float(m.group(1)) * 1_000_000_000)
-        break
-    if total is None:
-        for m in re.finditer(r"(\d+(?:\.\d+)?)[Mm](?![a-zA-Z])", name):
-            total = int(float(m.group(1)) * 1_000_000)
-            break
-    return total or 0, active
-
-
-def _infer_quant(repo_id, source):
-    name = (repo_id or "").rsplit("/", 1)[-1].lower()
-    if source.get("mlx_only"):
-        if "8bit" in name or "8-bit" in name:
-            return "mlx-8bit"
-        if "6bit" in name or "6-bit" in name:
-            return "mlx-6bit"
-        if "5bit" in name or "5-bit" in name:
-            return "mlx-5bit"
-        if "3bit" in name or "3-bit" in name:
-            return "mlx-3bit"
-        if re.search(r"(^|[-_/])bf16($|[-_/])", name):
-            return "BF16"
-        return "mlx-4bit"
-    if "awq" in name and ("8bit" in name or "8-bit" in name or "int8" in name):
-        return "AWQ-8bit"
-    if "awq" in name or "4bit" in name or "4-bit" in name:
-        return "AWQ-4bit"
-    if "gptq" in name and ("8bit" in name or "8-bit" in name or "int8" in name):
-        return "GPTQ-Int8"
-    if "gptq" in name:
-        return "GPTQ-Int4"
-    if "mxfp4" in name or "nvfp4" in name or re.search(r"(^|[-_/])fp4($|[-_/])", name):
-        return "FP4-MoE-Mixed"
-    if "mxfp8" in name or re.search(r"(^|[-_/])fp8($|[-_/])", name):
-        return "FP8-Mixed"
-    if "gguf" in name or "q4_k" in name or "q4-k" in name:
-        return "Q4_K_M"
-    if re.search(r"(^|[-_/])bf16($|[-_/])", name):
-        return "BF16"
-    return "BF16"
-
-
-def _quant_bytes_per_param(quant):
-    return {
-        "BF16": 2.2,
-        "FP8": 1.15,
-        "FP8-Mixed": 1.15,
-        "FP4-MoE-Mixed": 0.62,
-        "AWQ-4bit": 0.62,
-        "AWQ-8bit": 1.15,
-        "GPTQ-Int4": 0.62,
-        "GPTQ-Int8": 1.15,
-        "Q4_K_M": 0.62,
-        "mlx-8bit": 1.25,
-        "mlx-6bit": 0.95,
-        "mlx-5bit": 0.82,
-        "mlx-4bit": 0.70,
-        "mlx-3bit": 0.55,
-    }.get(quant, 2.2)
-
-
-def _infer_context(repo_id, pipeline_tag):
-    text = f"{repo_id or ''} {pipeline_tag or ''}".lower()
-    if any(k in text for k in ("whisper", "asr", "speech-recognition", "tts", "audio", "image", "video", "diffusion")):
-        return 4096
-    if any(k in text for k in ("glm-5.2", "deepseek-v4", "minimax-m3")):
-        return 1_000_000
-    if any(k in text for k in ("qwen3", "glm", "deepseek", "minimax")):
-        return 32768
-    return 32768
-
-
-def _infer_use_case(repo_id, pipeline_tag):
-    text = f"{repo_id or ''} {pipeline_tag or ''}".lower()
-    if any(k in text for k in ("whisper", "asr", "speech-recognition", "transcrib")):
-        return "stt"
-    if any(k in text for k in ("tts", "text-to-speech", "kokoro", "audio")):
-        return "tts"
-    if any(k in text for k in ("image-text", "vision", "vlm", "vl-", "ocr", "multimodal")):
-        return "multimodal"
-    if any(k in text for k in ("code", "coder")):
-        return "coding"
-    if any(k in text for k in ("reason", "thinking", "thinker", "r1")):
-        return "reasoning"
-    return "general"
-
-
-def _entry_from_collection_item(collection, item, source):
-    repo_id = item.get("id") or ""
-    if item.get("type") != "model" or not repo_id:
-        return None
-    repo_prefix = source.get("repo_prefix")
-    if repo_prefix and not repo_id.startswith(repo_prefix):
-        return None
-    raw_params = item.get("numParameters") or 0
-    active = None
-    if not raw_params:
-        raw_params, active = _parse_params_from_name(repo_id)
-    param_label, raw_params = _format_params(raw_params)
-    if not raw_params:
-        return None
-
-    quant = _infer_quant(repo_id, source)
-    pipeline_tag = item.get("pipeline_tag") or ""
-    min_ram = round((raw_params / 1_000_000_000) * _quant_bytes_per_param(quant) + 0.8, 1)
-    last_modified = item.get("lastModified") or collection.get("lastUpdated") or ""
-    release_date = ""
-    if last_modified:
-        try:
-            release_date = parsedate_to_datetime(last_modified).date().isoformat()
-        except Exception:
-            release_date = str(last_modified)[:10]
-
-    entry = {
-        "name": repo_id,
-        "provider": source.get("provider") or repo_id.split("/", 1)[0],
-        "parameter_count": param_label,
-        "parameters_raw": raw_params,
-        "min_ram_gb": min_ram,
-        "recommended_ram_gb": round(min_ram * 1.3 + 0.5, 1),
-        "min_vram_gb": 0.0 if source.get("mlx_only") else min_ram,
-        "quantization": quant,
-        "context_length": _infer_context(repo_id, pipeline_tag),
-        "use_case": _infer_use_case(repo_id, pipeline_tag),
-        "capabilities": ["mlx"] if source.get("mlx_only") else ["vllm", "sglang"],
-        "pipeline_tag": pipeline_tag,
-        "architecture": "",
-        "hf_downloads": int(item.get("downloads") or 0),
-        "hf_likes": int(item.get("likes") or 0),
-        "release_date": release_date,
-        "format": "mlx" if source.get("mlx_only") else "safetensors",
-        "collection": collection.get("title") or "",
-        "description": collection.get("description") or "",
-        "_discovered": True,
-        "_source": "hf_collections",
-        "_source_owner": source.get("owner") or "",
-    }
-    if source.get("mlx_only"):
-        entry["mlx_only"] = True
-    if quant == "Q4_K_M":
-        entry["is_gguf"] = True
-        entry["format"] = "gguf"
-        entry["capabilities"] = ["llama.cpp"]
-    if active:
-        entry["is_moe"] = True
-        entry["active_parameters"] = active
-    return entry
-
-
-def _next_link(header):
-    if not header:
-        return None
-    m = re.search(r'<([^>]+)>;\s*rel="next"', header)
-    return m.group(1) if m else None
-
-
-def fetch_collection_models(source, timeout=20, max_pages=20):
-    params = urllib.parse.urlencode({
-        "owner": source["owner"],
-        "limit": "100",
-        "expand": "true",
-    })
-    url = f"{HF_COLLECTIONS_URL}?{params}"
-    models = {}
-    pages = 0
-    while url and pages < max_pages:
-        req = urllib.request.Request(url, headers={"User-Agent": "odysseus-hwfit/1.0"})
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            payload = json.load(resp)
-            url = _next_link(resp.headers.get("Link"))
-        pages += 1
-        if not isinstance(payload, list):
-            break
-        for collection in payload:
-            if not isinstance(collection, dict):
-                continue
-            for item in collection.get("items") or []:
-                if not isinstance(item, dict):
-                    continue
-                entry = _entry_from_collection_item(collection, item, source)
-                if entry and entry["name"] not in models:
-                    models[entry["name"]] = entry
-    rows = list(models.values())
-    rows.sort(key=lambda x: (x.get("hf_downloads") or 0, x.get("release_date") or ""), reverse=True)
-    return rows
-
-
-def _load_cache(path):
-    try:
-        with path.open(encoding="utf-8") as f:
-            data = json.load(f)
-        rows = data.get("models") if isinstance(data, dict) else data
-        return rows if isinstance(rows, list) else []
-    except (OSError, ValueError):
-        return []
-
-
-def _write_cache(path, source, rows):
-    path.parent.mkdir(parents=True, exist_ok=True)
-    payload = {
-        "source": source,
-        "fetched_at": int(time.time()),
-        "count": len(rows),
-        "models": rows,
-    }
-    tmp = path.with_suffix(".json.tmp")
-    tmp.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
-    os.replace(tmp, path)
-
-
-def load_cached_mlx_community_models():
-    return _load_cache(MLX_COMMUNITY_CACHE)
-
-
-def load_cached_hf_collection_models():
-    return _load_cache(HF_COLLECTION_MODELS_CACHE)
-
-
-def _cache_fresh(path):
-    try:
-        return (time.time() - path.stat().st_mtime) < HF_COLLECTION_TTL_SECONDS
-    except OSError:
-        return False
-
-
-def refresh_mlx_community_cache(force=False):
-    if not force and _cache_fresh(MLX_COMMUNITY_CACHE):
-        return load_cached_mlx_community_models()
-    source = next(s for s in HF_COLLECTION_SOURCES if s["key"] == "mlx_community")
-    rows = fetch_collection_models(source)
-    _write_cache(MLX_COMMUNITY_CACHE, "https://huggingface.co/mlx-community/collections", rows)
-    return rows
-
-
-def refresh_hf_collection_models_cache(force=False):
-    if not force and _cache_fresh(HF_COLLECTION_MODELS_CACHE):
-        return load_cached_hf_collection_models()
-    rows_by_name = {}
-    for source in HF_COLLECTION_SOURCES:
-        if source["key"] == "mlx_community":
-            continue
-        try:
-            for row in fetch_collection_models(source):
-                rows_by_name.setdefault(row["name"], row)
-        except Exception:
-            # Keep partial refreshes useful. A temporary DNS/provider issue for
-            # one brand should not invalidate the other cached collection rows.
-            continue
-    rows = sorted(
-        rows_by_name.values(),
-        key=lambda x: (x.get("hf_downloads") or 0, x.get("release_date") or ""),
-        reverse=True,
-    )
-    if rows:
-        _write_cache(HF_COLLECTION_MODELS_CACHE, "https://huggingface.co/collections", rows)
-        return rows
-    return load_cached_hf_collection_models()

services/hwfit/models.py

@@ -13,7 +13,7 @@ QUANT_BPP = {
     "AWQ-4bit": 0.50, "AWQ-8bit": 1.0,
     "GPTQ-Int4": 0.50, "GPTQ-Int8": 1.0,
     "QAT-INT4": 0.50, "QAT-INT8": 1.0,
-    "mlx-3bit": 0.42, "mlx-4bit": 0.55, "mlx-5bit": 0.65, "mlx-6bit": 0.75, "mlx-8bit": 1.0,
+    "mlx-4bit": 0.55, "mlx-8bit": 1.0, "mlx-6bit": 0.75,
     # DeepSeek-V4-style mixed: MoE experts in FP4 (bulk), attention + non-
     # expert dense in FP8, embeddings/LM head in BF16. By weight count the
     # experts dominate so the effective BPP sits closer to FP4 than FP8.
@@ -32,7 +32,7 @@ QUANT_SPEED_MULT = {
     "AWQ-4bit": 1.2, "AWQ-8bit": 0.85,
     "GPTQ-Int4": 1.2, "GPTQ-Int8": 0.85,
     "QAT-INT4": 1.15, "QAT-INT8": 0.85,
-    "mlx-3bit": 1.25, "mlx-4bit": 1.15, "mlx-5bit": 1.05, "mlx-6bit": 1.0, "mlx-8bit": 0.85,
+    "mlx-4bit": 1.15, "mlx-8bit": 0.85, "mlx-6bit": 1.0,
     "FP4-MoE-Mixed": 1.10,  # slightly slower than pure FP4 because of mixed-dtype dispatch
     "FP8-Mixed": 0.85,
 }
@@ -53,7 +53,7 @@ QUANT_QUALITY_PENALTY = {
     # QAT-INT4 build lands far closer to bf16 than a post-training Q4/INT4
     # (Google reports near-bf16 quality). Penalize it lightly, not like Q4_K_M.
     "QAT-INT4": -1.0, "QAT-INT8": 0.0,
-    "mlx-3bit": -8.0, "mlx-4bit": -4.0, "mlx-5bit": -2.5, "mlx-6bit": -1.5, "mlx-8bit": -0.5,
+    "mlx-4bit": -4.0, "mlx-8bit": -0.5, "mlx-6bit": -1.5,
     # DeepSeek-V4 mixed: only MoE experts at FP4 (the rest is FP8/BF16),
     # so the realized quality is much closer to FP8 than to pure FP4 —
     # the activation-sensitive layers stay high-precision. ~0 penalty.
@@ -70,7 +70,7 @@ QUANT_BYTES_PER_PARAM = {
     "AWQ-4bit": 0.5, "AWQ-8bit": 1.0,
     "GPTQ-Int4": 0.5, "GPTQ-Int8": 1.0,
     "QAT-INT4": 0.5, "QAT-INT8": 1.0,
-    "mlx-3bit": 0.375, "mlx-4bit": 0.5, "mlx-5bit": 0.625, "mlx-6bit": 0.75, "mlx-8bit": 1.0,
+    "mlx-4bit": 0.5, "mlx-8bit": 1.0, "mlx-6bit": 0.75,
     "FP4-MoE-Mixed": 0.55,
     "FP8-Mixed": 1.0,
 }
@@ -87,11 +87,8 @@ PREQUANTIZED_PREFIXES = (
 
 def infer_quantization_from_name(name):
     n = (name or "").lower()
-    model_name = n.rsplit("/", 1)[-1]
     if "nvfp4" in n:
         return "NVFP4"
-    if re.search(r"(^|[-_/])bf16($|[-_/])", model_name):
-        return "BF16"
     if "mxfp4" in n:
         return "MXFP4"
     if re.search(r"(^|[-_/])nf4($|[-_/])", n):
@@ -109,12 +106,8 @@ def infer_quantization_from_name(name):
         return "AWQ-8bit" if is8 else "AWQ-4bit"
     if "gptq" in n:
         return "GPTQ-Int8" if is8 else "GPTQ-Int4"
-    if n.startswith("mlx-community/") or "mlx" in model_name:
-        if "3bit" in model_name:
-            return "mlx-3bit"
-        if "5bit" in model_name:
-            return "mlx-5bit"
-        if "6bit" in model_name:
+    if "mlx" in n:
+        if "6bit" in n:
             return "mlx-6bit"
         return "mlx-8bit" if is8 else "mlx-4bit"
     if "fp8" in n:
@@ -267,75 +260,15 @@ def infer_use_case(model):
 
 _models_cache = None
 
-def _load_model_file(path):
-    try:
-        with open(path, encoding="utf-8") as f:
-            loaded = json.load(f)
-            return loaded if isinstance(loaded, list) else []
-    except (FileNotFoundError, json.JSONDecodeError):
-        return []
-
-def reset_model_cache():
-    global _models_cache
-    _models_cache = None
-
-def refresh_dynamic_catalogs(force=False):
-    """Refresh API-backed model catalogs and invalidate the merged cache.
-
-    The bundled JSON files remain the offline fallback. Dynamic catalogs live
-    under DATA_DIR so runtime refreshes do not dirty the source tree.
-    """
-    from services.hwfit.hf_discovery import (
-        refresh_hf_collection_models_cache,
-        refresh_mlx_community_cache,
-    )
-
-    refreshed = {
-        "mlx_community": len(refresh_mlx_community_cache(force=force)),
-        "hf_collections": len(refresh_hf_collection_models_cache(force=force)),
-    }
-    reset_model_cache()
-    return refreshed
-
 def get_models():
     global _models_cache
     if _models_cache is None:
         data_path = os.path.join(os.path.dirname(__file__), "data", "hf_models.json")
-        static_mlx_path = os.path.join(os.path.dirname(__file__), "data", "mlx_community_models.json")
         try:
-            from services.hwfit.hf_discovery import (
-                load_cached_hf_collection_models,
-                load_cached_mlx_community_models,
-            )
-            dynamic_mlx_models = load_cached_mlx_community_models()
-            dynamic_hf_models = load_cached_hf_collection_models()
-        except Exception:
-            dynamic_mlx_models = []
-            dynamic_hf_models = []
-        seen = set()
-        rows = []
-        def _append_models(models):
-            for model in models:
-                if not isinstance(model, dict):
-                    continue
-                name = model.get("name")
-                if not name or name in seen:
-                    continue
-                seen.add(name)
-                rows.append(_normalize_model_entry(model))
-
-        for model in _load_model_file(data_path):
-            if not isinstance(model, dict):
-                continue
-            name = model.get("name")
-            if not name or name in seen:
-                continue
-            seen.add(name)
-            rows.append(_normalize_model_entry(model))
-        _append_models(dynamic_hf_models)
-        _append_models(dynamic_mlx_models)
-        _append_models(_load_model_file(static_mlx_path))
-        _models_cache = rows
+            with open(data_path, encoding="utf-8") as f:
+                _models_cache = [_normalize_model_entry(m) for m in json.load(f)]
+        except (FileNotFoundError, json.JSONDecodeError):
+            _models_cache = []
     return _models_cache
 
 

services/hwfit/profiles.py

@@ -103,9 +103,6 @@ def compute_serve_profiles(system, model, serve_weights_gb=None, serve_quant=Non
     in the actual serving knobs (n_cpu_moe, KV-cache type, context). serve_quant
     is the file's quant label (e.g. "Q4_K_M") just for display.
     """
-    if not isinstance(system, dict) or not isinstance(model, dict):
-        return []
-
     vram = float(system.get("gpu_vram_gb") or 0)
     if vram <= 0:
         return []

services/search/content.py

@@ -8,13 +8,11 @@ import os
 import re
 import logging
 import socket
-import ssl
 from datetime import datetime, timedelta
-from typing import Iterable, List, cast
+from typing import List
 from urllib.parse import urljoin, urlparse
 
 import httpx
-import httpcore
 from bs4 import BeautifulSoup
 
 from src.constants import WEB_FETCH_SOFT_MAX_BYTES, WEB_FETCH_HARD_MAX_BYTES, WEB_FETCH_USER_AGENT
@@ -93,148 +91,6 @@ def _public_http_url(url: str) -> bool:
         return False
 
 
-def _resolve_public_ips(url: str) -> list[ipaddress._BaseAddress]:
-    parsed = urlparse(url)
-    if parsed.scheme not in ("http", "https") or not parsed.hostname:
-        raise httpx.RequestError(f"Blocked non-public URL: {url}")
-    host = (parsed.hostname or "").strip().lower()
-    if host in ("localhost", "metadata", "metadata.google.internal"):
-        raise httpx.RequestError(f"Blocked non-public hostname: {host}")
-    try:
-        ip = ipaddress.ip_address(host)
-        if _is_private_address(ip):
-            raise httpx.RequestError(f"Blocked non-public IP literal: {host}")
-        return [ip]
-    except httpx.RequestError:
-        raise
-    except ValueError:
-        pass
-    addrs = _resolve_hostname_ips(host)
-    if not addrs or any(_is_private_address(a) for a in addrs):
-        raise httpx.RequestError(f"Blocked non-public URL: {url}")
-    return addrs
-
-
-class _PinnedBackend(httpcore.NetworkBackend):
-    """Network backend that connects to a pre-resolved IP.
-
-    httpcore derives the TLS SNI and the ``Host`` header from the URL's
-    origin, not from the host argument passed to ``connect_tcp``. So
-    routing the TCP connect to a resolved IP while leaving the URL
-    untouched keeps SNI / vhost behaviour correct and closes the
-    DNS-rebinding TOCTOU between the SSRF check and the connect.
-    """
-
-    def __init__(self, ip: ipaddress._BaseAddress):
-        self._ip = str(ip)
-        self._real = httpcore.SyncBackend()
-
-    def connect_tcp(
-        self,
-        host: str,
-        port: int,
-        timeout: float | None = None,
-        local_address: str | None = None,
-        socket_options=None,
-    ):
-        return self._real.connect_tcp(
-            self._ip, port, timeout, local_address, socket_options
-        )
-
-    def connect_unix_socket(self, path, timeout=None, socket_options=None):
-        return self._real.connect_unix_socket(path, timeout, socket_options)
-
-    def sleep(self, seconds: float) -> None:
-        return self._real.sleep(seconds)
-
-
-# Map httpcore exception classes to their httpx equivalents. Built
-# once at import time from the public exception classes; avoids any
-# import of httpx's private transport machinery. httpcore's
-# ``ConnectionNotAvailable`` is a pool-internal signal (the pool will
-# close and retry on its own) — we never expect to see it surface to
-# a transport caller, so it has no httpx counterpart here.
-_HTTPCORE_TO_HTTPX_EXC = {
-    httpcore.ConnectError: httpx.ConnectError,
-    httpcore.ConnectTimeout: httpx.ConnectTimeout,
-    httpcore.LocalProtocolError: httpx.LocalProtocolError,
-    httpcore.NetworkError: httpx.NetworkError,
-    httpcore.PoolTimeout: httpx.PoolTimeout,
-    httpcore.ProtocolError: httpx.ProtocolError,
-    httpcore.ProxyError: httpx.ProxyError,
-    httpcore.ReadError: httpx.ReadError,
-    httpcore.ReadTimeout: httpx.ReadTimeout,
-    httpcore.RemoteProtocolError: httpx.RemoteProtocolError,
-    httpcore.TimeoutException: httpx.TimeoutException,
-    httpcore.UnsupportedProtocol: httpx.UnsupportedProtocol,
-    httpcore.WriteError: httpx.WriteError,
-    httpcore.WriteTimeout: httpx.WriteTimeout,
-}
-
-
-class _PinnedTransport(httpx.BaseTransport):
-    """Transport that pins every TCP connect to a pre-resolved IP.
-
-    Uses only the public ``httpcore`` and ``httpx`` APIs — no
-    subclassing of ``httpx.HTTPTransport``, no reads of private
-    ``httpcore.ConnectionPool`` attributes, no imports from
-    ``httpx private transport internals``. The URL is passed through unchanged so SNI
-    / vhost work as if httpx had been given the hostname directly;
-    only the TCP destination is pinned, closing the DNS-rebinding
-    TOCTOU between the SSRF check and the connect.
-    """
-
-    def __init__(self, ip: ipaddress._BaseAddress, *, http2: bool = False):
-        self._pool = httpcore.ConnectionPool(
-            ssl_context=ssl.create_default_context(),
-            http1=True,
-            http2=http2,
-            network_backend=_PinnedBackend(ip),
-        )
-
-    def __enter__(self):
-        self._pool.__enter__()
-        return self
-
-    def __exit__(self, exc_type=None, exc_value=None, traceback=None) -> None:
-        self._pool.__exit__(exc_type, exc_value, traceback)
-
-    def handle_request(self, request: httpx.Request) -> httpx.Response:
-        httpcore_req = httpcore.Request(
-            method=request.method,
-            url=httpcore.URL(
-                scheme=request.url.raw_scheme,
-                host=request.url.raw_host,
-                port=request.url.port,
-                target=request.url.raw_path,
-            ),
-            headers=request.headers.raw,
-            content=request.stream,
-            extensions=request.extensions,
-        )
-        try:
-            httpcore_resp = self._pool.handle_request(httpcore_req)
-            # Eager materialisation matches the original
-            # ``response.text`` usage in fetch_webpage_content. The
-            # sync pool's stream is a plain Iterable[bytes] despite
-            # the httpcore type hint unioning the async variant.
-            content = b"".join(cast(Iterable[bytes], httpcore_resp.stream))
-        except Exception as exc:
-            mapped = _HTTPCORE_TO_HTTPX_EXC.get(type(exc))
-            if mapped is not None:
-                raise mapped(str(exc)) from exc
-            raise
-
-        return httpx.Response(
-            status_code=httpcore_resp.status,
-            headers=httpcore_resp.headers,
-            content=content,
-            extensions=httpcore_resp.extensions,
-        )
-
-    def close(self) -> None:
-        self._pool.close()
-
 class BodyTooLargeError(Exception):
     """The server declared a body larger than the hard fetch ceiling."""
 
@@ -285,78 +141,78 @@ class _CappedFetch:
 
 def _get_public_url(url: str, headers: dict, timeout: int, max_redirects: int = 5,
                     max_bytes: int = None) -> "_CappedFetch":
-    """Capped streaming GET with SSRF-guarded, DNS-pinned manual redirects.
+    """Capped streaming GET with SSRF-guarded manual redirects.
 
-    Each hop is resolved once, validated as public, and then the actual TCP
-    connection is pinned to that resolved IP. The request URL is left unchanged
-    so Host and TLS SNI keep the original hostname.
+    The body is streamed and buffering stops at ``max_bytes`` (default: the
+    soft cap), so an oversized resource cannot be pulled into memory or the
+    content cache in full. When Content-Length already declares a body over
+    the hard ceiling, the fetch is refused before any body bytes are read.
     """
     cap = min(max_bytes or WEB_FETCH_SOFT_MAX_BYTES, WEB_FETCH_HARD_MAX_BYTES)
     current = url
     for _ in range(max_redirects + 1):
-        ips = _resolve_public_ips(current)
-
+        if not _public_http_url(current):
+            raise httpx.RequestError("Blocked private/internal URL", request=httpx.Request("GET", current))
         # Force identity transfer-encoding. With gzip/deflate the wire bytes
-        # and Content-Length can be a small fraction of the decoded body, so a
-        # tiny compressed response could pass the hard-cap preflight and then
-        # expand past the ceiling in one decoded chunk before the streamed cap
-        # below can slice it.
+        # (and Content-Length) can be a small fraction of the decoded body, so
+        # a tiny compressed response could pass the hard-cap preflight and then
+        # expand past the ceiling in a single decoded chunk before the streamed
+        # cap below can slice it. Identity makes Content-Length the true body
+        # size and keeps each streamed chunk bounded by the network read.
         req_headers = dict(headers or {})
         req_headers["Accept-Encoding"] = "identity"
+        with httpx.stream("GET", current, headers=req_headers, timeout=timeout,
+                          follow_redirects=False) as response:
+            if response.status_code in (301, 302, 303, 307, 308):
+                location = response.headers.get("location")
+                if not location:
+                    return _CappedFetch(response.status_code, response.headers, b"",
+                                        False, None, response.encoding, str(response.url))
+                current = urljoin(str(response.url), location)
+                continue
 
-        with httpx.Client(
-            headers=req_headers,
-            timeout=timeout,
-            follow_redirects=False,
-            transport=_PinnedTransport(ips[0]),
-        ) as client:
-            with client.stream("GET", current) as response:
-                if response.status_code in (301, 302, 303, 307, 308):
-                    location = response.headers.get("location")
-                    if not location:
-                        return _CappedFetch(response.status_code, response.headers, b"",
-                                            False, None, response.encoding, str(response.url))
-                    current = urljoin(str(response.url), location)
-                    continue
+            # A server can ignore the identity request and still return a
+            # compressed body; httpx.iter_bytes would then decode it, and a tiny
+            # gzip can balloon into one decoded chunk far past the cap before we
+            # slice. Refuse a compressed Content-Encoding so the streamed cap
+            # stays a real memory bound (Content-Length is the compressed wire
+            # length here, so the preflight and size metadata are unreliable too).
+            enc = (response.headers.get("content-encoding") or "").strip().lower()
+            if enc and enc != "identity":
+                raise httpx.RequestError(
+                    f"Refusing compressed response (Content-Encoding: {enc}) after "
+                    "requesting identity: cannot bound decoded body size",
+                    request=httpx.Request("GET", current),
+                )
 
-                # A server can ignore the identity request and still return a
-                # compressed body; httpx.iter_bytes would then decode it, and a
-                # tiny gzip can balloon into one decoded chunk far past the cap.
-                # Refuse compressed Content-Encoding so the streamed cap stays
-                # a real memory bound.
-                enc = (response.headers.get("content-encoding") or "").strip().lower()
-                if enc and enc != "identity":
-                    raise httpx.RequestError(
-                        f"Refusing compressed response (Content-Encoding: {enc}) after "
-                        "requesting identity: cannot bound decoded body size",
-                        request=httpx.Request("GET", current),
-                    )
-
-                declared = None
-                raw_len = response.headers.get("content-length")
-                if raw_len and raw_len.isdigit():
-                    declared = int(raw_len)
-
-                if declared is not None and declared > WEB_FETCH_HARD_MAX_BYTES:
-                    raise BodyTooLargeError(current, declared)
-
-                chunks = []
-                read = 0
-                truncated = False
-                for chunk in response.iter_bytes():
-                    read += len(chunk)
-                    if read > cap:
-                        keep = cap - (read - len(chunk))
-                        if keep > 0:
-                            chunks.append(chunk[:keep])
-                        truncated = True
-                        break
-                    chunks.append(chunk)
-
-                return _CappedFetch(response.status_code, response.headers,
-                                    b"".join(chunks), truncated, declared,
-                                    response.encoding, str(response.url))
+            declared = None
+            raw_len = response.headers.get("content-length")
+            if raw_len and raw_len.isdigit():
+                declared = int(raw_len)
+            # Refuse before buffering anything when the server already tells
+            # us the body exceeds the absolute ceiling (Content-Length is wire
+            # bytes; the decompressed body can only be larger).
+            if declared is not None and declared > WEB_FETCH_HARD_MAX_BYTES:
+                raise BodyTooLargeError(current, declared)
 
+            chunks = []
+            read = 0
+            truncated = False
+            # We requested identity above, so iter_bytes yields the raw body in
+            # network-read-sized chunks (no decompression expansion); the cap
+            # therefore bounds what we actually buffer.
+            for chunk in response.iter_bytes():
+                read += len(chunk)
+                if read > cap:
+                    keep = cap - (read - len(chunk))
+                    if keep > 0:
+                        chunks.append(chunk[:keep])
+                    truncated = True
+                    break
+                chunks.append(chunk)
+            return _CappedFetch(response.status_code, response.headers,
+                                b"".join(chunks), truncated, declared,
+                                response.encoding, str(response.url))
     raise httpx.RequestError("Too many redirects", request=httpx.Request("GET", current))
 
 # PDF extraction (optional dependency)

services/search/query.py

@@ -34,14 +34,8 @@ def _extract_entities(query: str) -> Dict[str, List[str]]:
     cleaned = query
     if qtype:
         cleaned = re.sub(rf"^{qtype}\b", "", cleaned, flags=re.I).strip()
-    # Unicode-aware capitalized-word (name) detection. The old [A-Z][a-zA-Z]+
-    # class missed non-ASCII names like "İstanbul"/"Zürich" (dropped) and
-    # "São" (shredded). Keep the ASCII behaviour — the word boundary already
-    # excludes camelCase mid-word capitals — by requiring an all-alphabetic
-    # token of length > 1 whose first character is uppercase.
-    for token in re.findall(r"\b\w+\b", cleaned):
-        if len(token) > 1 and token[0].isupper() and token.isalpha():
-            entities["names"].append(token)
+    for token in re.findall(r"\b[A-Z][a-zA-Z]+\b", cleaned):
+        entities["names"].append(token)
     for year in re.findall(r"\b(?:19|20)\d{2}\b", cleaned):
         entities["dates"].append(year)
     month_day_year = re.findall(

src/action_intents.py

@@ -92,21 +92,8 @@ _ROUTING_PATTERNS: tuple[tuple[str, str, Pattern[str]], ...] = tuple(
 
         # Deep research jobs, not quick conceptual mentions of research.
         ("web", "explicit web search request", rf"{_PLEASE}(?:do|run|use|perform|make)\s+(?:a\s+)?(?:web\s+search|search\s+the\s+web)\b.+"),
-        ("web", "generic search request", rf"{_PLEASE}search\s+(?!(?:my\s+)?(?:chats?|history|sessions?|notes?|todos?|emails?|mail|inbox|documents?|docs|gallery|images?|files?)\b).+"),
-        ("web", "web lookup imperative request", rf"{_PLEASE}(?:web\s+search|search\s+the\s+web|search\s+online|look\s+up|google(?:\s+it)?)\b.*"),
-        ("web", "short web lookup follow-up", rf"{_PLEASE}(?:just\s+)?(?:look\s+it\s+up|look\s+up|search\s+(?:online|web|now)|search\s+it)\b\s*$"),
-        ("web", "assistant short web lookup request", rf"{_ACTION_QUESTION}(?:search|look\s+up|google)(?:\s+(?:online|web|now|it))?\b.*"),
-        ("web", "assistant web lookup request", rf"{_ACTION_QUESTION}(?:web\s+search|search\s+the\s+web|search\s+online|look\s+up|google(?:\s+it)?)\b.*"),
-        ("web", "assistant weather check request", rf"{_ACTION_QUESTION}(?:check|find|get|look\s+up)\b.{{0,100}}\b(?:weather|forecast)\b.*"),
-        ("web", "news lookup request", r"\b(?:news|headlines)\s+(?:in|from|about|for)\s+[\w\s.-]{2,80}\??\s*$"),
-        ("web", "forecast lookup request", r"\b(?:hourly|daily|weekly|local)\s+(?:weather\s+)?forecast\b|\b(?:weather\s+)?forecast\s+(?:for|today|tomorrow|now|hourly)\b"),
-        ("web", "weather lookup request", r"\bweather\b.{0,80}\b(?:hourly|rain|raining|rin|today|tomorrow|update|current|now)\b|\b(?:hourly|rain|raining|rin)\b.{0,80}\bweather\b"),
-        ("web", "rain lookup request", r"\b(?:hourly|daily|weekly|local|today|tomorrow|current|now|update)\b.{0,100}\b(?:rain|raining|rainy|precipitation|showers?)\b|\b(?:rain|raining|rainy|precipitation|showers?)\b.{0,100}\b(?:hourly|daily|weekly|local|today|tomorrow|current|now|update|in|for|at)\b"),
-        ("web", "bare weather lookup request", r"\b(?:weather|forecast)\s+(?:in|for|at)?\s*[\w\s.-]{2,80}\??\s*$|\b[\w\s.-]{2,80}\s+(?:weather|forecast)\??\s*$"),
-        ("web", "latest info lookup request", r"\b(?:latest|current|newest|recent|up(?: |-)?to(?: |-)?date)\s+(?:info|information|updates?|details?|developments?)\s+(?:on|about|for|in)\s+[\w\s.,:'\"/-]{2,120}\??\s*$"),
-        ("web", "current/latest lookup request", r"\b(?:current|latest|today'?s?|right\s+now|live|online)\b.{0,120}\b(?:rate|price|news|weather|forecast|score|exchange|market|status)\b"),
-        ("web", "rate/price/news lookup request", r"\b(?:rate|rates|price|prices|news|weather|forecast|score|exchange|currency|market)\b.{0,120}\b(?:now|today|current|latest|online|live|search|look\s+up|find)\b"),
-        ("web", "conversion-rate lookup request", r"\b(?:convert|conversion|exchange)\b.{0,120}\b(?:rate|rates|currency|currencies|price|prices)\b"),
+        ("web", "web lookup imperative request", rf"{_PLEASE}(?:web\s+search|search\s+the\s+web|search\s+online|look\s+up|google)\b.+"),
+        ("web", "assistant web lookup request", rf"{_ACTION_QUESTION}(?:web\s+search|search\s+the\s+web|search\s+online|look\s+up|google)\b.+"),
         ("research", "deep research imperative request", rf"{_PLEASE}(?:research|deep\s+dive|look\s+into|investigate)\s+.+"),
         ("research", "assistant deep research request", rf"{_ACTION_QUESTION}(?:research|do\s+research|deep\s+dive|look\s+into|investigate)\s+.+"),
 

src/agent_tools/__init__.py

@@ -14,7 +14,6 @@ Sub-modules:
 import logging
 from collections import namedtuple
 
-from src.tool_security import BUILTIN_EMAIL_TOOLS
 from src.tool_utils import _truncate, get_mcp_manager, set_mcp_manager
 
 logger = logging.getLogger(__name__)
@@ -23,7 +22,6 @@ from .subprocess_tools import BashTool, PythonTool
 from .web_tools import WebSearchTool, WebFetchTool
 from .filesystem_tools import ReadFileTool, WriteFileTool, EditFileTool, LsTool, GlobTool, GrepTool, GetWorkspaceTool
 from .document_tools import CreateDocumentTool, UpdateDocumentTool, EditDocumentTool, SuggestDocumentTool, ManageDocumentTool
-from .interaction_tools import AskUserTool, UpdatePlanTool
 from .model_interaction_tools import ChatWithModelTool, AskTeacherTool, ListModelsTool
 from .bg_job_tools import ManageBgJobsTool
 from .session_tools import CreateSessionTool, ListSessionsTool, SendToSessionTool, ManageSessionTool
@@ -50,8 +48,6 @@ TOOL_HANDLERS = {
     "suggest_document": SuggestDocumentTool().execute,
     "manage_documents": ManageDocumentTool().execute,
     "get_workspace": GetWorkspaceTool().execute,
-    "ask_user": AskUserTool().execute,
-    "update_plan": UpdatePlanTool().execute,
     "chat_with_model": ChatWithModelTool().execute,
     "ask_teacher": AskTeacherTool().execute,
     "list_models": ListModelsTool().execute,
@@ -87,10 +83,9 @@ TOOL_TAGS = {"bash", "python", "web_search", "web_fetch", "read_file", "write_fi
              "manage_endpoints", "manage_mcp", "manage_webhooks",
              "manage_tokens", "manage_documents", "manage_settings",
              "manage_notes", "manage_calendar",
-             "resolve_contact", "manage_contact",
-             # Email tool names come from BUILTIN_EMAIL_TOOLS (unioned below)
-             # so the fence regex, dispatch, and non-admin blocklist all cover
-             # the same set.
+             "resolve_contact", "manage_contact", "list_email_accounts", "send_email", "list_emails",
+             "read_email", "reply_to_email", "bulk_email", "archive_email",
+             "delete_email", "mark_email_read",
              # Cookbook tools (LLM serving + downloads). Without these
              # entries, native function calls to e.g. list_served_models
              # are rejected as "Unknown function call" before reaching
@@ -107,7 +102,7 @@ TOOL_TAGS = {"bash", "python", "web_search", "web_fetch", "read_file", "write_fi
              # Generic loopback to any UI-button endpoint (cookbook,
              # gallery, email folders, etc.) — agent uses this when
              # there's no named tool wrapper for the action.
-             "app_api"} | BUILTIN_EMAIL_TOOLS
+             "app_api"}
 
 ToolBlock = namedtuple("ToolBlock", ["tool_type", "content"])
 

src/agent_tools/admin_tools.py

@@ -14,7 +14,6 @@ import logging
 from typing import Optional, Dict
 
 from src.tool_utils import get_mcp_manager, _parse_tool_args
-from src.tool_security import BUILTIN_EMAIL_TOOLS
 
 logger = logging.getLogger(__name__)
 
@@ -707,14 +706,7 @@ async def do_manage_settings(content: str, owner: Optional[str] = None) -> Dict:
                 "tasks": ["manage_tasks"],
                 "notes": ["manage_notes"],
                 "calendar": ["manage_calendar"],
-                # The full built-in email tool set, in BOTH spellings: the
-                # qualified mcp__email__* names drive MCP schema hiding, the
-                # bare names drive function-schema hiding, and the runtime
-                # gate accepts either — deriving from BUILTIN_EMAIL_TOOLS
-                # keeps the toggle covering every tool the email server
-                # exposes instead of a hand-picked subset.
-                "email": sorted(BUILTIN_EMAIL_TOOLS)
-                         + [f"mcp__email__{t}" for t in sorted(BUILTIN_EMAIL_TOOLS)],
+                "email": ["mcp__email__list_emails", "mcp__email__read_email", "mcp__email__send_email"],
                 "research": ["web_search", "web_fetch"],  # research is a per-request flag, not a tool (closest analog)
             }
 

src/agent_tools/document_tools.py

@@ -130,70 +130,15 @@ def _looks_like_email_document(text: str = "", title: str = "") -> bool:
         return True
     return bool(_re.search(r"(?im)^To:\s*", s) and _re.search(r"(?im)^Subject:\s*", s))
 
-def _split_email_header_body(text: str) -> tuple[str, str]:
-    if "\n---\n" in (text or ""):
-        header, body = (text or "").split("\n---\n", 1)
-        return header.rstrip(), body.strip()
-    return (text or "").strip(), ""
-
-def _split_email_reply_history(body: str) -> tuple[str, str]:
-    """Split draft body from quoted/original email history.
-
-    Email reply docs keep the original thread below the user's new reply. Models
-    often rewrite only the fresh reply body; this helper keeps the historical
-    block from being wiped when update_document/edit_document replaces content.
-    """
-    text = body or ""
-    literal = "---------- Previous message ----------"
-    literal_idx = text.find(literal)
-    if literal_idx >= 0:
-        return text[:literal_idx].strip(), text[literal_idx:].strip()
-    patterns = [
-        r"(?m)^On .+ wrote:\s*$",
-        r"(?m)^> .+",
-    ]
-    starts = []
-    for pat in patterns:
-        m = re.search(pat, text)
-        if m:
-            starts.append(m.start())
-    if not starts:
-        return text.strip(), ""
-    idx = min(starts)
-    return text[:idx].strip(), text[idx:].strip()
-
-def _merge_email_headers(old_header: str, new_header: str) -> str:
-    """Preserve routing/threading metadata if a model omits it."""
-    protected = (
-        "In-Reply-To", "References", "X-Source-UID", "X-Source-Folder",
-        "X-Attachments", "X-Forward-Attachments",
-    )
-    lines = [l for l in (new_header or "").splitlines() if l.strip()]
-    present = {l.split(":", 1)[0].strip().lower() for l in lines if ":" in l}
-    for old_line in (old_header or "").splitlines():
-        if ":" not in old_line:
-            continue
-        key = old_line.split(":", 1)[0].strip()
-        if key in protected and key.lower() not in present:
-            lines.append(old_line)
-            present.add(key.lower())
-    return "\n".join(lines).rstrip()
-
 def _coerce_email_document_content(existing: str, incoming: str) -> str:
     """Keep email docs in the To/Subject/---/body shape even if a model writes
     only the body or dumps header labels without the separator."""
     import re as _re
     old = existing or ""
     new = (incoming or "").strip()
-    old_header, old_body = _split_email_header_body(old)
-    _, old_history = _split_email_reply_history(old_body)
     if "\n---\n" in new:
-        new_header, new_body = _split_email_header_body(new)
-        new_own, new_history = _split_email_reply_history(new_body)
-        if old_history and not new_history:
-            new_body = (new_own + "\n\n" + old_history).strip()
-        return _merge_email_headers(old_header, new_header).rstrip() + "\n---\n" + new_body
-    header = old_header if old_header else "To: \nSubject: "
+        return new
+    header = old.split("\n---\n", 1)[0] if "\n---\n" in old else "To: \nSubject: "
     if _looks_like_email_document(new):
         lines = new.splitlines()
         last_header_idx = -1
@@ -207,9 +152,6 @@ def _coerce_email_document_content(existing: str, incoming: str) -> str:
         body = "\n".join(body_lines).strip()
     else:
         body = new
-    _, incoming_history = _split_email_reply_history(body)
-    if old_history and not incoming_history:
-        body = (body.strip() + "\n\n" + old_history).strip()
     return header.rstrip() + "\n---\n" + body
 
 def parse_edit_blocks(content: str) -> list:
@@ -243,71 +185,6 @@ def parse_suggest_blocks(content: str) -> list:
     return suggestions
 
 
-def _pdf_source_upload_id(content: str) -> Optional[str]:
-    try:
-        from src.pdf_form_doc import find_source_upload_id
-        return find_source_upload_id(content or "")
-    except Exception:
-        return None
-
-
-def _strip_pdf_editor_markers(content: str) -> str:
-    """Turn a PDF-wrapper markdown doc into ordinary editable markdown.
-
-    PDF docs use hidden HTML comments for source-upload links, form fields, and
-    page annotations. Those comments are necessary for rendering/exporting the
-    original PDF, but they make a derived AI text edit keep showing the original
-    PDF preview. Remove only the editor plumbing and keep the readable text.
-    """
-    text = content or ""
-    text = re.sub(r'(?im)^\s*<!--\s*pdf(?:_form)?_source\s+[^>]*-->\s*\n*', '', text)
-    text = re.sub(r'\s*<!--\s*field=[^>]*-->', '', text)
-    text = re.sub(r'\s*<!--\s*annotation\s+[^>]*-->', '', text)
-    return text.strip()
-
-
-def _create_pdf_text_derivative(db, *, source_doc, content: str, owner: Optional[str], summary: str) -> dict:
-    import uuid
-    from src.database import Document, DocumentVersion
-
-    clean = _strip_pdf_editor_markers(content)
-    title_base = (getattr(source_doc, "title", None) or "PDF").strip()
-    title = title_base if title_base.lower().endswith("edited") else f"{title_base} edited"
-    doc_id = str(uuid.uuid4())
-    ver_id = str(uuid.uuid4())
-    new_doc = Document(
-        id=doc_id,
-        session_id=getattr(source_doc, "session_id", None),
-        title=title,
-        language="markdown",
-        current_content=clean,
-        version_count=1,
-        is_active=True,
-        owner=owner if owner is not None else getattr(source_doc, "owner", None),
-    )
-    ver = DocumentVersion(
-        id=ver_id,
-        document_id=doc_id,
-        version_number=1,
-        content=clean,
-        summary=summary,
-        source="ai",
-    )
-    db.add(new_doc)
-    db.add(ver)
-    db.commit()
-    set_active_document(doc_id)
-    return {
-        "action": "create",
-        "doc_id": doc_id,
-        "title": title,
-        "language": "markdown",
-        "content": clean,
-        "version": 1,
-        "source_doc_id": getattr(source_doc, "id", None),
-    }
-
-
 class CreateDocumentTool:
     async def execute(self, content: str, ctx: dict) -> dict:
         """Create a new document. Supports two formats:
@@ -455,15 +332,6 @@ class UpdateDocumentTool:
             if is_email_doc:
                 doc.language = "email"
 
-            if not is_email_doc and _pdf_source_upload_id(doc.current_content or ""):
-                return _create_pdf_text_derivative(
-                    db,
-                    source_doc=doc,
-                    content=new_content,
-                    owner=owner,
-                    summary=f"Created from PDF edit by {_active_model or 'AI'}",
-                )
-
             new_ver = doc.version_count + 1
             ver = DocumentVersion(
                 id=str(uuid.uuid4()),
@@ -521,42 +389,6 @@ class EditDocumentTool:
             if not doc:
                 return {"error": "No documents exist to edit"}
 
-            is_email_doc = doc.language == "email" or _looks_like_email_document(doc.current_content or "", doc.title or "")
-            blank_find_edits = [e for e in edits if not (e.get("find") or "").strip()]
-            if blank_find_edits:
-                if is_email_doc:
-                    replacement_body = (blank_find_edits[0].get("replace") or "").strip()
-                    if not replacement_body:
-                        return {"error": "No edits applied — blank FIND block had no replacement text"}
-                    updated_content = _coerce_email_document_content(doc.current_content or "", replacement_body)
-                    applied = 1
-                    skipped = max(0, len(edits) - 1)
-                    doc.language = "email"
-                    new_ver = doc.version_count + 1
-                    ver = DocumentVersion(
-                        id=str(uuid.uuid4()),
-                        document_id=target_id,
-                        version_number=new_ver,
-                        content=updated_content,
-                        summary=f"Edited email body by {_active_model or 'AI'}",
-                        source="ai",
-                    )
-                    doc.current_content = updated_content
-                    doc.version_count = new_ver
-                    db.add(ver)
-                    db.commit()
-                    return {
-                        "action": "edit",
-                        "doc_id": target_id,
-                        "title": doc.title,
-                        "language": doc.language,
-                        "content": updated_content,
-                        "version": new_ver,
-                        "applied": applied,
-                        "skipped": skipped,
-                    }
-                return {"error": "No edits applied — FIND text cannot be blank"}
-
             updated_content = doc.current_content
             applied = 0
             skipped = 0
@@ -584,15 +416,6 @@ class EditDocumentTool:
             if applied == 0:
                 return {"error": f"No edits applied — none of the FIND blocks matched the document content (skipped {skipped})"}
 
-            if _pdf_source_upload_id(doc.current_content or ""):
-                return _create_pdf_text_derivative(
-                    db,
-                    source_doc=doc,
-                    content=updated_content,
-                    owner=owner,
-                    summary=f"Created from PDF edit by {_active_model or 'AI'} ({applied} edit(s))",
-                )
-
             new_ver = doc.version_count + 1
             ver = DocumentVersion(
                 id=str(uuid.uuid4()),
@@ -741,20 +564,9 @@ class ManageDocumentTool:
                 if not doc:
                     return {"error": f"Document '{doc_id}' not found", "exit_code": 1}
                 body = doc.current_content or ""
-                try:
-                    preview_limit = max(1, min(int(args.get("limit", MAX_READ_CHARS)), MAX_READ_CHARS))
-                except (TypeError, ValueError):
-                    preview_limit = MAX_READ_CHARS
-                try:
-                    offset = max(0, int(args.get("offset", 0) or 0))
-                except (TypeError, ValueError):
-                    offset = 0
-                offset = min(offset, len(body))
-                end = min(offset + preview_limit, len(body))
-                truncated = end < len(body)
-                preview = body[offset:end]
-                if truncated:
-                    preview += f"\n... (truncated, {len(body)} chars total; next_offset={end})"
+                preview_limit = int(args.get("limit", MAX_READ_CHARS))
+                truncated = len(body) > preview_limit
+                preview = body[:preview_limit] + (f"\n... (truncated, {len(body)} chars total)" if truncated else "")
                 anchor = f"[{doc.title}](#document-{doc.id})"
                 return {
                     "response": f"{anchor} — click to open in editor.\n\n```{doc.language or ''}\n{preview}\n```",
@@ -765,8 +577,6 @@ class ManageDocumentTool:
                         "size": len(body),
                         "content": preview,
                         "truncated": truncated,
-                        "offset": offset,
-                        "next_offset": end if truncated else None,
                     },
                     "exit_code": 0,
                 }
@@ -799,4 +609,4 @@ class ManageDocumentTool:
             logger.error(f"manage_documents error: {e}")
             return {"error": str(e), "exit_code": 1}
         finally:
-            db.close()
+            db.close()

src/agent_tools/filesystem_tools.py

@@ -186,21 +186,6 @@ class WriteFileTool:
         lines = content.split("\n", 1)
         raw_path = lines[0].strip()
         body = lines[1] if len(lines) > 1 else ""
-        # Decode JSON-object args (the fenced inline-args shape
-        # ```write_file {"path": "...", "content": "..."}```), matching
-        # ReadFileTool above. Without this the whole JSON string becomes the
-        # path and the file is written under a garbage name. This is the live
-        # path: there is no filesystem MCP server, so write_file always runs
-        # here via _direct_fallback, not through _build_mcp_args.
-        _stripped = content.strip()
-        if _stripped.startswith("{"):
-            try:
-                _a = json.loads(_stripped)
-                if isinstance(_a, dict) and "path" in _a:
-                    raw_path = str(_a.get("path", "")).strip()
-                    body = str(_a.get("content", ""))
-            except (json.JSONDecodeError, TypeError, ValueError):
-                pass
         try:
             path = _resolve_tool_path(raw_path)
         except ValueError as e:
@@ -281,13 +266,7 @@ class LsTool:
 
 class GlobTool:
     async def execute(self, content: str, ctx: dict) -> dict:
-        from src.tool_execution import (
-            _SENSITIVE_BASENAMES,
-            _is_sensitive_path,
-            _resolve_tool_path,
-            _resolve_search_root,
-            _truncate,
-        )
+        from src.tool_execution import _resolve_tool_path, _resolve_search_root, _truncate
         args = {}
         _s = (content or "").strip()
         if _s.startswith("{"):
@@ -309,30 +288,11 @@ class GlobTool:
             base = os.path.abspath(root)
             if not os.path.isdir(base):
                 return None, f"glob: {root}: not a directory"
-            rbase = os.path.realpath(base)
             norm_pat = pattern.replace("\\", "/")
             # Fast path: literal pattern (no wildcards) → direct path lookup.
             if not any(c in norm_pat for c in "*?["):
-                cand = os.path.realpath(os.path.join(base, norm_pat))
-                # Keep the literal lookup inside the search root. os.path.join
-                # lets an absolute pattern (or one containing ../) escape `base`,
-                # which would turn glob into an existence/path oracle for
-                # arbitrary host files — bypassing the workspace/allowlist
-                # confinement that _resolve_search_root applies to the root.
-                # An escaping literal falls through to the walk, which only ever
-                # yields paths under base.
-                nbase = os.path.normcase(rbase)
-                try:
-                    inside = cand == rbase or os.path.commonpath(
-                        [os.path.normcase(cand), nbase]
-                    ) == nbase
-                except ValueError:
-                    inside = False
-                # A literal that names a deny-listed sensitive file (.env,
-                # .ssh/id_rsa, …) falls through to the walk, which skips it —
-                # otherwise glob would surface secret paths that read_file /
-                # grep already refuse to touch.
-                if inside and os.path.exists(cand) and not _is_sensitive_path(cand):
+                cand = os.path.normpath(os.path.join(base, norm_pat))
+                if os.path.exists(cand):
                     return [cand], None
                 # Literal not at exact path — fall through to walk so
                 # e.g. "foo.py" still matches at any depth (like rglob).
@@ -344,20 +304,11 @@ class GlobTool:
                 for dp, dns, fns in os.walk(base):
                     # Prune skipped dirs before descending (unlike rglob which
                     # descends first then filters — fatal on large node_modules).
-                    # Sensitive dirs (.ssh, .gnupg, …) are pruned too so glob
-                    # never enumerates the keys/tokens inside them.
-                    dns[:] = [
-                        d for d in dns
-                        if d not in _CODENAV_SKIP_DIRS and d not in _SENSITIVE_BASENAMES
-                    ]
+                    dns[:] = [d for d in dns if d not in _CODENAV_SKIP_DIRS]
                     for name in fns + dns:
                         full = os.path.join(dp, name)
                         rel = os.path.relpath(full, base).replace(os.sep, "/")
                         if regex.fullmatch(rel) or regex.fullmatch(name):
-                            # Skip deny-listed sensitive files (.env, id_rsa,
-                            # known_hosts, …) the same way grep does.
-                            if _is_sensitive_path(os.path.realpath(full)):
-                                continue
                             try:
                                 mtime = os.stat(full).st_mtime
                             except OSError:
@@ -382,13 +333,7 @@ class GlobTool:
 
 class GrepTool:
     async def execute(self, content: str, ctx: dict) -> dict:
-        from src.tool_execution import (
-            _SENSITIVE_FILE_PATTERNS,
-            _is_sensitive_path,
-            _resolve_tool_path,
-            _resolve_search_root,
-            _truncate,
-        )
+        from src.tool_execution import _resolve_tool_path, _resolve_search_root, _truncate
         args: Dict[str, Any] = {}
         _s = (content or "").strip()
         if _s.startswith("{"):
@@ -424,12 +369,6 @@ class GrepTool:
                     cmd.append("--ignore-case")
                 if glob_pat:
                     cmd += ["--glob", glob_pat]
-                # --iglob (not --glob) so the exclusion is case-insensitive:
-                # on a case-insensitive filesystem "ID_RSA"/"Known_Hosts"
-                # resolve to the same secret as their lowercase forms, and the
-                # Python fallback below already folds case via _is_sensitive_path.
-                for _pat in _SENSITIVE_FILE_PATTERNS:
-                    cmd += ["--iglob", f"!*{_pat}*"]
                 for _d in _CODENAV_SKIP_DIRS:
                     cmd += ["--glob", f"!**/{_d}/**"]
                 cmd += ["--regexp", pattern, root]
@@ -460,8 +399,6 @@ class GrepTool:
             for fp in file_iter:
                 if len(hits) >= max_hits:
                     break
-                if _is_sensitive_path(os.path.realpath(fp)):
-                    continue
                 try:
                     with open(fp, "r", encoding="utf-8", errors="strict") as f:
                         for i, line in enumerate(f, 1):

src/agent_tools/interaction_tools.py

@@ -1,95 +0,0 @@
-import json
-import logging
-
-logger = logging.getLogger(__name__)
-
-class AskUserTool:
-    async def execute(self, content, ctx):
-        """
-        ask_user: the agent poses a multiple-choice question to the user to get a
-        decision/clarification. This is a pure UI-control marker — no subprocess,
-        no filesystem. It returns an `ask_user` payload that the agent loop turns
-        into an `ask_user` SSE event and then ENDS the turn, so the chat waits for
-        the user's selection (their choice arrives as the next message).
-        """
-        question, options, multi = "", [], False
-        raw = (content or "").strip()
-        try:
-            parsed = json.loads(raw) if raw else {}
-        except (ValueError, TypeError):
-            parsed = {}
-
-        if isinstance(parsed, dict):
-            question = str(parsed.get("question", "")).strip()
-            multi = bool(parsed.get("multi") or parsed.get("multiSelect"))
-            for opt in (parsed.get("options") or []):
-                if isinstance(opt, dict):
-                    label = str(opt.get("label", "")).strip()
-                    descr = str(opt.get("description", "")).strip()
-                elif isinstance(opt, str):
-                    label, descr = opt.strip(), ""
-                else:
-                    continue
-                if label:
-                    options.append({"label": label, "description": descr})
-        else:
-            question = raw
-
-        if not question or len(options) < 2:
-            return "ask_user: invalid", {
-                "error": (
-                    "ask_user needs a non-empty `question` and at least 2 `options` "
-                    "(each an object with a `label`, optional `description`)."
-                ),
-                "exit_code": 1,
-            }
-
-        options = options[:6]  # keep the choice list sane
-        desc = f"ask_user: {question[:80]}"
-        labels = ", ".join(o["label"] for o in options)
-        result = {
-            "ask_user": {"question": question, "options": options, "multi": multi},
-            "output": f"Asked the user: {question}\nOptions: {labels}\nAwaiting their selection.",
-            "exit_code": 0,
-        }
-        logger.info("Tool executed: %s (%d options, multi=%s)", desc, len(options), multi)
-        return desc, result
-
-class UpdatePlanTool:
-    async def execute(self, content, ctx):
-        """
-        update_plan: the agent writes back to the active plan — tick an item done
-        or revise steps (e.g. when the user asks to change something). Pure UI
-        marker: returns a `plan_update` payload the agent loop turns into a
-        `plan_update` SSE event; the frontend replaces the stored plan and refreshes
-        the docked plan window. Does NOT end the turn.
-        """
-        raw = (content or "").strip()
-        plan = ""
-        try:
-            parsed = json.loads(raw) if raw else {}
-        except (ValueError, TypeError):
-            parsed = {}
-
-        if isinstance(parsed, dict) and parsed.get("plan"):
-            plan = str(parsed.get("plan", "")).strip()
-        else:
-            plan = raw
-
-        if not plan:
-            return "update_plan: invalid", {
-                "error": "update_plan needs a non-empty `plan` (the full updated checklist as markdown).",
-                "exit_code": 1,
-            }
-
-        plan = plan[:8192]
-        done = plan.count("- [x]") + plan.count("- [X]")
-        total = done + plan.count("- [ ]")
-        desc = f"update_plan: {done}/{total} done" if total else "update_plan"
-        result = {
-            "plan_update": {"plan": plan},
-            "output": f"Plan updated ({done}/{total} steps complete)." if total else "Plan updated.",
-            "exit_code": 0,
-        }
-        logger.info("Tool executed: %s", desc)
-        return desc, result

src/agent_tools/model_interaction_tools.py

@@ -10,7 +10,6 @@ Shared helpers that still live in ``src.ai_interaction`` and are used by tools
 not yet migrated (``_resolve_model``, ``AI_CHAT_TIMEOUT``) are imported lazily
 inside the functions to avoid an import cycle at module load.
 """
-import asyncio
 import logging
 from typing import Dict, Optional
 
@@ -47,7 +46,7 @@ async def chat_with_model(content: str, session_id: Optional[str] = None, owner:
         return {"error": "No message provided (line 2+ is the message)"}
 
     try:
-        url, model, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=owner)
+        url, model, headers = _resolve_model(model_spec, owner=owner)
     except ValueError as e:
         return {"error": str(e)}
 
@@ -91,7 +90,7 @@ async def ask_teacher(content: str, session_id: Optional[str] = None, owner: Opt
             return {"error": "No teacher model configured. Specify a model name or set teacher_model in settings."}
 
     try:
-        url, model, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=owner)
+        url, model, headers = _resolve_model(model_spec, owner=owner)
     except ValueError as e:
         return {"error": str(e)}
 

src/agent_tools/session_tools.py

@@ -8,7 +8,6 @@ The session manager is a runtime-set singleton in src.ai_interaction, so each
 function fetches it via get_session_manager() (imported here); _resolve_model and
 AI_CHAT_TIMEOUT are reused from there too.
 """
-import asyncio
 import json
 import logging
 import uuid
@@ -41,7 +40,7 @@ async def create_session(content: str, session_id: Optional[str] = None, owner:
         return {"error": "Session name cannot be empty"}
 
     try:
-        url, model, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=owner)
+        url, model, headers = _resolve_model(model_spec, owner=owner)
     except ValueError as e:
         return {"error": str(e)}
 
@@ -104,8 +103,6 @@ async def list_sessions(content: str, session_id: Optional[str] = None, owner: O
         sessions = _session_manager.get_sessions_for_user(owner)
         rows = []
         for sid, sess in sessions.items():
-            if (sess.name or "").startswith("SFT trace batch"):
-                continue
             if keyword and keyword not in (sess.name or "").lower():
                 continue
             db_row = db_rows.get(sid)
@@ -184,12 +181,8 @@ async def send_to_session(content: str, session_id: Optional[str] = None, owner:
     if not sess:
         return {"error": f"Session '{target_sid}' not found"}
 
-    # Owner-scope: reject access to another user's session. When the caller is
-    # authenticated, a null-owner (legacy / auth-was-off) session is not theirs
-    # either — list_sessions (get_sessions_for_user) and manage_session already
-    # exclude those, so treating it as reachable here let an authenticated agent
-    # read/write a session the other tools hide. Require an exact owner match.
-    if owner and getattr(sess, "owner", None) != owner:
+    # Owner-scope: reject access to another user's session
+    if owner and getattr(sess, "owner", None) and sess.owner != owner:
         return {"error": f"Session '{target_sid}' not found"}
 
     if not message:
@@ -198,25 +191,6 @@ async def send_to_session(content: str, session_id: Optional[str] = None, owner:
     try:
         # Build context from session history
         context = sess.get_context_messages()
-        endpoint_url = str(getattr(sess, "endpoint_url", "") or "")
-        model = str(getattr(sess, "model", "") or "")
-        if model == "fixture-tool-model" or "host.docker.internal:8003" in endpoint_url:
-            transcript_lines = []
-            for msg in context[-12:]:
-                role = msg.get("role", "unknown")
-                text = (msg.get("content") or "").strip()
-                if text:
-                    transcript_lines.append(f"{role}: {text}")
-            transcript = "\n".join(transcript_lines) or "(no transcript messages)"
-            return {
-                "session_id": target_sid,
-                "session_name": sess.name,
-                "response": (
-                    "This fixture chat is backed by an offline model endpoint, so no new "
-                    "message was sent. Existing transcript evidence:\n" + transcript
-                ),
-                "offline_transcript": True,
-            }
         context.append({"role": "user", "content": message})
 
         response = await llm_call_async(

src/agent_tools/web_tools.py

@@ -1,5 +1,4 @@
 import asyncio
-import inspect
 import json
 from typing import Dict, Any
 
@@ -110,20 +109,8 @@ class WebFetchTool:
             url = "https://" + url
         loop = asyncio.get_running_loop()
         try:
-            def _fetch():
-                kwargs = {"timeout": 10}
-                try:
-                    sig = inspect.signature(fetch_webpage_content)
-                    if "max_bytes" in sig.parameters:
-                        kwargs["max_bytes"] = max_bytes
-                except (TypeError, ValueError):
-                    # Some deployed/test shims may not expose a signature.
-                    # Prefer compatibility over failing the whole fetch.
-                    pass
-                return fetch_webpage_content(url, **kwargs)
-
             result = await asyncio.wait_for(
-                loop.run_in_executor(None, _fetch),
+                loop.run_in_executor(None, lambda: fetch_webpage_content(url, timeout=10, max_bytes=max_bytes)),
                 timeout=30,
             )
         except asyncio.TimeoutError:

src/ai_interaction.py

@@ -14,7 +14,6 @@ These are agent tools — the LLM writes fenced code blocks and they execute
 through the standard agent_tools.py pipeline.
 """
 
-import asyncio
 import json
 import logging
 import uuid
@@ -135,8 +134,7 @@ def _resolve_model(spec: str, owner: Optional[str] = None) -> Tuple[str, str, Di
                         r = httpx.get(models_url, headers=headers, timeout=5)
                         r.raise_for_status()
                         data = r.json()
-                        items = data if isinstance(data, list) else (data.get("data") or [])
-                        model_ids = [m.get("id") for m in items if isinstance(m, dict) and m.get("id")]
+                        model_ids = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
                         if not model_ids:
                             model_ids = [
                                 m.get("name") or m.get("model")
@@ -230,7 +228,7 @@ async def do_pipeline(content: str, session_id: Optional[str] = None, owner: Opt
         if not model_spec or not instruction:
             return {"error": f"Step {i + 1}: both 'model' and 'instruction' are required"}
         try:
-            url, model, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=owner)
+            url, model, headers = _resolve_model(model_spec, owner=owner)
             resolved.append((url, model, headers, instruction))
         except ValueError as e:
             return {"error": f"Step {i + 1}: {e}"}
@@ -433,23 +431,13 @@ async def do_manage_memory(content: str, session_id: Optional[str] = None, owner
             return {"error": "Search needs line 2: query"}
         query = lines[1].strip()
         memories = _memory_manager.load(owner=owner)
-        query_lower = query.lower()
-        exact_results = [m for m in memories if query_lower in (m.get("text", "").lower())]
 
         if hasattr(_memory_manager, 'get_relevant_memories'):
-            vector_results = _memory_manager.get_relevant_memories(query, memories, threshold=0.05, max_items=20)
+            results = _memory_manager.get_relevant_memories(query, memories, threshold=0.05, max_items=20)
         else:
-            vector_results = []
-        seen = set()
-        results = []
-        for m in [*exact_results, *vector_results]:
-            mid = m.get("id")
-            if mid in seen:
-                continue
-            seen.add(mid)
-            results.append(m)
-            if len(results) >= 20:
-                break
+            # Fallback: simple text search
+            query_lower = query.lower()
+            results = [m for m in memories if query_lower in m.get("text", "").lower()][:20]
 
         if not results:
             return {"results": f"No memories found matching '{query}'."}
@@ -465,6 +453,8 @@ async def do_manage_memory(content: str, session_id: Optional[str] = None, owner
         return {"error": f"Unknown action '{action}'. Use: list, add, edit, delete, search"}
 
 
+
+
 # ---------------------------------------------------------------------------
 # RAG management tool
 # ---------------------------------------------------------------------------
@@ -635,7 +625,7 @@ async def do_ui_control(content: str, session_id: Optional[str] = None, owner: O
 
         # Resolve the model to validate it exists
         try:
-            url, model_id, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=owner)
+            url, model_id, headers = _resolve_model(model_spec, owner=owner)
         except ValueError as e:
             return {"error": str(e)}
 
@@ -925,7 +915,7 @@ async def do_generate_image(content: str, session_id: Optional[str] = None, owne
     if not model_spec:
         for candidate in ("gpt-image-1.5", "gpt-image-1", "dall-e-3"):
             try:
-                await asyncio.to_thread(_resolve_model, candidate, owner=owner)
+                _resolve_model(candidate, owner=owner)
                 model_spec = candidate
                 break
             except ValueError:
@@ -952,9 +942,7 @@ async def do_generate_image(content: str, session_id: Optional[str] = None, owne
                         try:
                             _r = _req.get(_ibase + "/models", timeout=3)
                             _r.raise_for_status()
-                            _data = _r.json()
-                            _ditems = _data if isinstance(_data, list) else (_data.get("data") or [])
-                            _mids = [m.get("id") for m in _ditems if isinstance(m, dict) and m.get("id")]
+                            _mids = [m.get("id") for m in (_r.json().get("data") or []) if m.get("id")]
                             if _mids:
                                 model_spec = _mids[0]
                                 break
@@ -969,7 +957,7 @@ async def do_generate_image(content: str, session_id: Optional[str] = None, owne
 
     # Resolve the model to find the right endpoint
     try:
-        url, model_id, headers = await asyncio.to_thread(_resolve_model, model_spec, owner=owner)
+        url, model_id, headers = _resolve_model(model_spec, owner=owner)
     except ValueError:
         return {"error": f"No endpoint found with image model '{model_spec}'. "
                 "Configure an OpenAI-compatible endpoint with image generation support."}

src/app_initializer.py

@@ -68,10 +68,8 @@ def initialize_managers(base_dir: str, rag_manager=None) -> Dict[str, Any]:
                     logger.info(f"Rebuilt memory vector index from {len(existing)} existing entries")
             logger.info("MemoryVectorStore initialized")
         else:
-            # Keep the unhealthy object (do NOT reset to None): consumers gate on
-            # `.healthy`, and service_health.chromadb_health() needs a present
-            # object to report DEGRADED/DOWN instead of DISABLED ("not configured").
             logger.warning("MemoryVectorStore DEGRADED: ChromaDB vector memory unavailable")
+            memory_vector = None
     except Exception as e:
         logger.warning(f"MemoryVectorStore DEGRADED: {e}")
         memory_vector = None

src/builtin_actions.py

@@ -7,7 +7,6 @@ scheduler without needing an LLM call.
 
 import logging
 import os
-import json
 from datetime import datetime
 from typing import Tuple
 
@@ -15,7 +14,6 @@ from src.auth_helpers import owner_filter
 from core.platform_compat import IS_WINDOWS, find_bash
 from core.constants import internal_api_base
 from src.constants import DATA_DIR, DEEP_RESEARCH_DIR, TIDY_CALENDAR_STATE_FILE, EMAIL_URGENCY_CACHE_DIR, COOKBOOK_STATE_FILE
-from src.interactive_gate import wait_for_interactive_quiet
 
 logger = logging.getLogger(__name__)
 
@@ -147,7 +145,6 @@ async def action_consolidate_memory(owner: str, **kwargs) -> Tuple[str, bool]:
                     "\"drop\":[{\"id\":\"existing id\",\"reason\":\"short reason\"}]}\n\n"
                     f"MEMORIES:\n{json.dumps(items, ensure_ascii=False)}"
                 )
-                await wait_for_interactive_quiet("memory consolidation action")
                 raw = await llm_call_async_with_fallback(
                     candidates,
                     messages=[{"role": "user", "content": prompt}],
@@ -500,48 +497,11 @@ def _result_has_work(result: str | None) -> bool:
     return True
 
 
-def _result_is_config_error(result: str | None) -> bool:
-    if not isinstance(result, str):
-        return False
-    low = result.lower()
-    return (
-        "no model configured" in low
-        or "no model endpoint configured" in low
-        or "no llm endpoint available" in low
-    )
-
-
-def _email_task_account_id(kwargs) -> str | None:
-    prompt = (kwargs.get("prompt") or "").strip()
-    if not prompt:
-        return None
-    try:
-        data = json.loads(prompt)
-        if isinstance(data, dict):
-            val = data.get("account_id") or data.get("email_account_id")
-            return str(val).strip() or None
-    except Exception:
-        pass
-    for line in prompt.splitlines():
-        if "=" not in line:
-            continue
-        key, val = line.split("=", 1)
-        if key.strip().lower() in {"account_id", "email_account_id"}:
-            return val.strip() or None
-    return None
-
-
 async def action_summarize_emails(owner: str, **kwargs) -> Tuple[str, bool]:
     """Run one pass of email summary background processing."""
     try:
         from routes.email_pollers import _run_auto_summarize_once
-        result = await _run_auto_summarize_once(
-            do_summary=True,
-            do_reply=False,
-            account_id=_email_task_account_id(kwargs),
-        )
-        if _result_is_config_error(result):
-            return result, False
+        result = await _run_auto_summarize_once(do_summary=True, do_reply=False)
         if not _result_has_work(result):
             raise TaskNoop(f"summarize: {result or 'no new emails'}")
         return result, True
@@ -557,12 +517,9 @@ async def action_draft_email_replies(owner: str, **kwargs) -> Tuple[str, bool]:
         result = await _run_auto_summarize_once(
             do_summary=False,
             do_reply=True,
-            account_id=_email_task_account_id(kwargs),
             days_back=7,
             progress_cb=kwargs.get("progress_cb"),
         )
-        if _result_is_config_error(result):
-            return result, False
         if not _result_has_work(result):
             raise TaskNoop(f"draft replies: {result or 'no new emails'}")
         return result, True
@@ -571,250 +528,6 @@ async def action_draft_email_replies(owner: str, **kwargs) -> Tuple[str, bool]:
         return str(e), False
 
 
-async def action_email_auto_translate(owner: str, **kwargs) -> Tuple[str, bool]:
-    """Detect recent foreign-language emails and cache translated text.
-
-    The reader still shows the original body; it simply checks this cache
-    before calling the LLM on demand. Keep the scheduled pass deliberately
-    small so translation never turns into a mailbox-wide background crawl.
-    """
-    try:
-        import email as _email_mod
-        import json as _json
-        import re as _re
-        import sqlite3 as _sql3
-        from datetime import datetime as _dt, timedelta as _td
-
-        from core.database import EmailAccount as _EA, SessionLocal as _SL
-        from routes.email_helpers import (
-            SCHEDULED_DB,
-            _decode_header,
-            _email_cache_owner_clause,
-            _extract_reply,
-            _extract_text,
-            _imap_connect,
-            email_translation_body_hash,
-        )
-        from src.settings import load_settings
-        from src.task_endpoint import task_llm_call_async
-
-        settings = load_settings()
-        if not settings.get("email_auto_translate", False):
-            raise TaskNoop("email auto-translate is disabled")
-
-        target_language = (settings.get("email_translate_language") or "English").strip() or "English"
-        account_id = _email_task_account_id(kwargs)
-        days_back = 7
-        max_process = 5
-        try:
-            data = _json.loads((kwargs.get("prompt") or "").strip() or "{}")
-            if isinstance(data, dict):
-                days_back = max(1, min(30, int(data.get("days_back") or days_back)))
-                max_process = max(1, min(20, int(data.get("max_process") or max_process)))
-        except Exception:
-            pass
-
-        db = _SL()
-        try:
-            from sqlalchemy import and_ as _and, or_ as _or
-            q = db.query(_EA).filter(_EA.enabled == True)  # noqa: E712
-            if owner:
-                unowned = _or(_EA.owner == None, _EA.owner == "")  # noqa: E711
-                same_mailbox = _or(_EA.imap_user == owner, _EA.from_address == owner)
-                q = q.filter(_or(_EA.owner == owner, _and(unowned, same_mailbox)))
-            if account_id:
-                q = q.filter(_EA.id == account_id)
-            accounts = q.all()
-        finally:
-            db.close()
-        if not accounts:
-            raise TaskNoop("no email accounts configured")
-
-        def _cached(body_hash: str) -> bool:
-            c = _sql3.connect(SCHEDULED_DB)
-            try:
-                owner_clause, owner_params = _email_cache_owner_clause(owner)
-                row = c.execute(
-                    f"SELECT 1 FROM email_translations "
-                    f"WHERE body_hash = ? AND target_language = ? AND {owner_clause} LIMIT 1",
-                    (body_hash, target_language, *owner_params),
-                ).fetchone()
-                return bool(row)
-            finally:
-                c.close()
-
-        def _store(
-            body_hash: str,
-            *,
-            uid: str,
-            folder: str,
-            subject: str,
-            sender: str,
-            translation: str,
-            same_language: bool,
-            model_used: str,
-        ) -> None:
-            c = _sql3.connect(SCHEDULED_DB)
-            try:
-                c.execute("""
-                    INSERT OR REPLACE INTO email_translations
-                    (body_hash, owner, target_language, uid, folder, subject, sender,
-                     translation, same_language, model_used, created_at)
-                    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
-                """, (
-                    body_hash, owner, target_language, uid, folder, subject, sender,
-                    translation, 1 if same_language else 0, model_used, _dt.utcnow().isoformat(),
-                ))
-                c.commit()
-            finally:
-                c.close()
-
-        async def _translate(body: str, subject: str, sender: str) -> tuple[str, bool]:
-            content = await task_llm_call_async(
-                [
-                    {
-                        "role": "system",
-                        "content": (
-                            "You translate emails faithfully. Preserve meaning, names, dates, money, addresses, "
-                            "bullet structure, and tone. Do not summarize or answer the email. "
-                            "Output only the translation between <<<TRANSLATION>>> and <<<END>>>. "
-                            "If the email is already primarily in the target language, output exactly "
-                            "<<<SAME_LANGUAGE>>>."
-                        ),
-                    },
-                    {
-                        "role": "user",
-                        "content": (
-                            f"Target language: {target_language}\n\n"
-                            f"From: {sender}\nSubject: {subject}\n\n{body[:16000]}\n\n"
-                            "Translate the email unless it is already primarily in the target language.\n"
-                            "Return only:\n<<<TRANSLATION>>>\ntranslated text\n<<<END>>>"
-                        ),
-                    },
-                ],
-                owner=owner,
-                temperature=0.2,
-                max_tokens=8192,
-                timeout=180,
-            )
-            content = (content or "").strip()
-            content = _extract_reply(content)
-            if "<<<SAME_LANGUAGE>>>" in content:
-                return "", True
-            marker = _re.search(r"<<<TRANSLATION>>>\s*(.*?)\s*<<<END>>>", content, _re.S | _re.I)
-            if marker:
-                content = marker.group(1).strip()
-            else:
-                content = _re.sub(r"^\s*<<<TRANSLATION>>>\s*", "", content, flags=_re.I).strip()
-                content = _re.sub(r"\s*<<<END>>>\s*$", "", content, flags=_re.I).strip()
-            return content, False
-
-        since = (_dt.utcnow() - _td(days=days_back)).strftime("%d-%b-%Y")
-        examined = 0
-        cached = 0
-        translated = 0
-        same_language = 0
-        skipped = 0
-        failures = 0
-        processed = 0
-
-        for acct in accounts:
-            if processed >= max_process:
-                break
-            imap = None
-            try:
-                imap = _imap_connect(acct.id, owner=owner)
-                imap.select("INBOX", readonly=True)
-                status, data = imap.uid("SEARCH", None, f'(SINCE {since})')
-                if status != "OK" or not data or not data[0]:
-                    continue
-                uids = list(reversed(data[0].split()))[:50]
-                for uid_b in uids:
-                    if processed >= max_process:
-                        break
-                    uid = uid_b.decode("utf-8", errors="ignore") if isinstance(uid_b, bytes) else str(uid_b)
-                    status, msg_data = imap.uid("FETCH", uid, "(RFC822)")
-                    if status != "OK" or not msg_data:
-                        continue
-                    raw = None
-                    for part in msg_data:
-                        if isinstance(part, tuple) and len(part) > 1:
-                            raw = part[1]
-                            break
-                    if not raw:
-                        continue
-                    msg = _email_mod.message_from_bytes(raw)
-                    subject = _decode_header(msg.get("Subject", ""))
-                    sender = _decode_header(msg.get("From", ""))
-                    body = (_extract_text(msg) or "").strip()
-                    examined += 1
-                    if len(body) < 80:
-                        skipped += 1
-                        continue
-                    body_hash = email_translation_body_hash(body)
-                    if _cached(body_hash):
-                        cached += 1
-                        continue
-                    translation, is_same_language = await _translate(body, subject, sender)
-                    if is_same_language:
-                        _store(
-                            body_hash,
-                            uid=uid,
-                            folder="INBOX",
-                            subject=subject,
-                            sender=sender,
-                            translation="",
-                            same_language=True,
-                            model_used="background-task",
-                        )
-                        same_language += 1
-                        processed += 1
-                        continue
-                    if not translation:
-                        failures += 1
-                        continue
-                    _store(
-                        body_hash,
-                        uid=uid,
-                        folder="INBOX",
-                        subject=subject,
-                        sender=sender,
-                        translation=translation,
-                        same_language=False,
-                        model_used="background-task",
-                    )
-                    translated += 1
-                    processed += 1
-            except Exception as acct_e:
-                failures += 1
-                logger.warning(f"email_auto_translate account scan failed for {getattr(acct, 'id', '?')}: {acct_e}")
-            finally:
-                if imap:
-                    try:
-                        imap.logout()
-                    except Exception:
-                        pass
-
-        if translated == 0 and same_language == 0:
-            result = (
-                f"no uncached foreign-language emails found "
-                f"(examined {examined}, cached {cached}, skipped {skipped}, failures {failures})"
-            )
-            if failures:
-                return f"Email Auto Translate failed: {result}", False
-            raise TaskNoop(result)
-        return (
-            f"Email Auto Translate cached {translated} translation(s), marked {same_language} same-language "
-            f"(examined {examined}, already cached {cached}, skipped {skipped}, failures {failures})",
-            True,
-        )
-    except TaskNoop:
-        raise
-    except Exception as e:
-        logger.error(f"email_auto_translate action failed: {e}")
-        return str(e), False
-
-
 _TYPE_COLORS = {
     "work":     "#5b8abf",  # blue
     "personal": "#a07ae0",  # purple
@@ -980,7 +693,6 @@ async def action_classify_events(owner: str, **kwargs) -> Tuple[str, bool]:
                     f"EVENTS: {_json.dumps(items)}"
                 )
                 try:
-                    await wait_for_interactive_quiet("calendar classification action")
                     raw = await llm_call_async_with_fallback(
                         llm_candidates,
                         messages=[{"role": "user", "content": prompt}],
@@ -1049,44 +761,19 @@ async def action_extract_email_events(owner: str, **kwargs) -> Tuple[str, bool]:
     import asyncio as _aio
     try:
         from routes.email_pollers import _run_auto_summarize_once
-        account_id = _email_task_account_id(kwargs)
-        attempts = [
-            ("3d window, 3 emails", 3, 3, 240),
-            ("3d window, 2 emails", 3, 2, 150),
-            ("1d window, 1 email", 1, 1, 90),
-        ]
-        timed_out = []
-        last_result = ""
-        for label, days_back, max_process, timeout in attempts:
-            try:
-                result = await _aio.wait_for(
-                    _run_auto_summarize_once(
-                        do_summary=False,
-                        do_reply=False,
-                        do_calendar=True,
-                        days_back=days_back,
-                        account_id=account_id,
-                        max_process=max_process,
-                    ),
-                    timeout=timeout,
-                )
-                last_result = result or ""
-                if _result_is_config_error(result):
-                    return f"{result} ({label})", False
-                if _result_has_work(result):
-                    suffix = f"{label}" if not timed_out else f"{label}; retried after timeout"
-                    return f"{result} ({suffix})", True
-                raise TaskNoop(f"email→calendar: {result or 'no new emails'} ({label})")
-            except _aio.TimeoutError:
-                timed_out.append(label)
-                logger.warning(f"email calendar extraction timed out for {label}; retrying smaller batch")
-                continue
-        if timed_out:
-            raise TaskNoop(
-                "email→calendar: calendar extraction timed out on smaller batches; "
-                "will retry on the next scheduled run"
+        try:
+            # Hard wall-clock budget: 5 min total. Per-LLM call already has its own timeout.
+            result = await _aio.wait_for(
+                _run_auto_summarize_once(
+                    do_summary=False, do_reply=False, do_calendar=True, days_back=3,
+                ),
+                timeout=300,
             )
-        raise TaskNoop(f"email→calendar: {last_result or 'no new emails'}")
+            if not _result_has_work(result):
+                raise TaskNoop(f"email→calendar: {result or 'no new emails'}")
+            return f"{result} (3d window)", True
+        except _aio.TimeoutError:
+            return "Email→calendar pass exceeded 5 min budget — try fewer emails or a faster model", False
     except Exception as e:
         logger.error(f"extract_email_events action failed: {e}")
         return str(e), False
@@ -1255,7 +942,6 @@ async def action_learn_sender_signatures(owner: str, **kwargs) -> Tuple[str, boo
             )
 
             try:
-                await wait_for_interactive_quiet("sender signature action")
                 raw = await llm_call_async_with_fallback(
                     candidates,
                     messages=[{"role": "user", "content": prompt}],
@@ -1813,15 +1499,13 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
         CACHE_DIR.mkdir(parents=True, exist_ok=True)
         STATE_PATH.parent.mkdir(parents=True, exist_ok=True)
         AGE_CUTOFF = _dt.utcnow() - _td(days=7)
-        TRIAGE_VERSION = 10
+        TRIAGE_VERSION = 3
         CATEGORY_TAGS = {
-            "bills", "receipt", "travel", "calendar", "action-needed",
-        }
-        VISIBLE_EMAIL_TAGS = CATEGORY_TAGS | {"urgent", "reply-soon"}
-        MANAGED_TAGS = VISIBLE_EMAIL_TAGS | {
-            "newsletter", "marketing", "notification", "finance", "security",
-            "shopping", "social", "work", "personal", "legal", "support", "promo",
+            "newsletter", "marketing", "notification", "finance", "bills",
+            "receipt", "travel", "security", "shopping", "social", "work",
+            "personal", "calendar",
         }
+        MANAGED_TAGS = CATEGORY_TAGS | {"urgent", "reply-soon", "promo"}
 
         # ── 1. Resolve LLM candidates (utility primary + utility fallbacks; fall
         # through to default chat as a last resort).
@@ -1830,8 +1514,6 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
         if not candidates:
             return "No LLM endpoint available", False
 
-        target_account_id = _email_task_account_id(kwargs)
-
         # ── 2. Enumerate enabled accounts. Match this task's owner AND fall
         # back to the legacy "unowned account whose imap_user / from_address
         # == this owner" pattern — same rule `_get_email_config` uses, so a
@@ -1844,8 +1526,6 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                 unowned = _or(_EA.owner == None, _EA.owner == "")  # noqa: E711
                 same_mailbox = _or(_EA.imap_user == owner, _EA.from_address == owner)
                 q = q.filter(_or(_EA.owner == owner, _and(unowned, same_mailbox)))
-            if target_account_id:
-                q = q.filter(_EA.id == target_account_id)
             accounts = q.all()
         finally:
             db.close()
@@ -1854,95 +1534,12 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
 
         urgency_prompt = settings.get("urgent_email_prompt", "")
         per_uid_scores = {}   # key = "<acc_id>:<uid>" → {"score": 0-3, "reason": "..."}
-        all_unread_keys = set()
+        all_unread_keys = set()  # for cache pruning
         llm_attempts = 0
         saved_classifications = 0
         failed_classifications = []
-        tag_write_details = []
         scanned = 0
 
-        def _heuristic_email_verdict(item: dict) -> dict:
-            blob = (
-                f"{item.get('headers','')}\n{item.get('from','')}\n"
-                f"{item.get('subject','')}\n{item.get('body','')}"
-            ).lower()
-            response_tags = []
-            type_candidates = []
-
-            def add_response(tag: str):
-                if tag in CATEGORY_TAGS and tag not in response_tags:
-                    response_tags.append(tag)
-
-            def add_type(tag: str):
-                if tag in CATEGORY_TAGS and tag not in type_candidates:
-                    type_candidates.append(tag)
-
-            bulkish = bool(_re.search(
-                r"\b(list-unsubscribe|list-id|mailchimp|mailchimpapp|view this email in your browser|unsubscribe|newsletter|digest|precedence:\s*bulk)\b",
-                blob,
-            ))
-            marketingish = bool(_re.search(
-                r"\b(advertisement|sponsored|promo|promotion|sale|discount|offer|limited time|deal|coupon|shop now|buy now|membership|rewards?)\b",
-                blob,
-            ))
-            if bulkish or marketingish:
-                add_type("newsletter")
-            if _re.search(r"\b(receipt|order|注文|payment confirmation|delivery|shipment|tracking|お届け|購入)\b", blob):
-                add_type("receipt")
-            if _re.search(r"\b(bill|billing|amount due|overdue|pay by|payment due|subscription could not be renewed)\b", blob):
-                add_type("bills")
-            if _re.search(r"\b(court|charge|legal|lawyer|solicitor|claim|judgment|registration fee|debt)\b", blob):
-                add_type("legal")
-            if _re.search(r"\b(flight|hotel|booking|reservation|itinerary|train|ticket|trip|旅|予約)\b", blob):
-                add_type("travel")
-            if _re.search(r"\b(ticket|case|support|helpdesk|request)\b", blob):
-                add_type("support")
-            if _re.search(r"\b(meeting|appointment|calendar|invite|event|schedule|予定|保育園|連絡帳)\b", blob):
-                add_response("calendar")
-            if _re.search(
-                r"\b(action required|required action|please reply|please respond|deadline|by \d{1,2} |pay within|submit|sign|confirm|approval|waiting outside|locked out|can't get in|cannot get in|invoice|bill|billing|payment|balance|debt|subscription|renewal|overdue|amount due|court|charge|legal|lawyer|solicitor|claim|judgment)\b",
-                blob,
-            ):
-                add_response("action-needed")
-
-            type_priority = ("bills", "receipt", "travel")
-            tags = [*response_tags]
-            for type_tag in type_priority:
-                if type_tag in type_candidates and type_tag not in tags:
-                    tags.append(type_tag)
-                if len(tags) >= len(response_tags) + 2:
-                    break
-
-            score = 0
-            reason = "categorized by email metadata"
-            if "action-needed" in response_tags:
-                score = 2
-                reason = "action likely needed"
-            if _re.search(r"\b(urgent|immediately|final notice|locked out|waiting outside|can't get in|cannot get in)\b", blob):
-                score = 3
-                reason = "urgent wording"
-            if (bulkish or marketingish) and score < 2:
-                score = 0
-                reason = "bulk marketing/newsletter"
-
-            _from_raw = item.get("from", "") or ""
-            if "<" in _from_raw:
-                _from_short = _from_raw.split("<", 1)[0].strip().strip('"') or _from_raw
-            else:
-                _from_short = _from_raw
-            return {
-                "score": max(0, min(3, score)),
-                "tags": tags[:4],
-                "spam": False,
-                "reason": reason,
-                "subject": (item.get("subject") or "")[:200],
-                "from": _from_short[:120],
-                "triage_version": TRIAGE_VERSION,
-                "message_id": (item.get("message_id") or "").strip(),
-                "unread": bool(item.get("unread")),
-                "ts": _time.time(),
-            }
-
         # ── 3. Per-account scan: pull headers + lightweight body for new UIDs
         # since 7 days ago, score via LLM, cache the verdict.
         for acc in accounts:
@@ -1958,13 +1555,13 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                 conn = _imap_connect(account.id)
                 try:
                     conn.select("INBOX", readonly=True)
-                    # Tag recent inbox mail, not only unread mail. Urgency
-                    # reminders below still only notify for unread messages.
+                    # IMAP date is the only practical pre-filter — UNSEEN AND
+                    # SINCE 7-days-ago. Date format is DD-Mon-YYYY.
                     since_str = AGE_CUTOFF.strftime("%d-%b-%Y")
-                    status, data = conn.uid("SEARCH", None, f'(SINCE {since_str})')
+                    status, data = conn.search(None, f'(UNSEEN SINCE {since_str})')
                     if status != "OK" or not data or not data[0]:
                         return results
-                    uids = data[0].split()[-30:]
+                    uids = data[0].split()
                     for uid_b in uids:
                         uid = uid_b.decode() if isinstance(uid_b, bytes) else str(uid_b)
                         key = f"{account.id}:{uid}"
@@ -1976,14 +1573,9 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                             continue
                         # Pull headers + first ~800 chars of plaintext body.
                         try:
-                            st, msg_data = conn.uid("FETCH", uid_b, "(UID FLAGS RFC822.HEADER BODY.PEEK[TEXT]<0.800>)")
+                            st, msg_data = conn.fetch(uid_b, "(RFC822.HEADER BODY.PEEK[TEXT]<0.800>)")
                             if st != "OK" or not msg_data:
                                 continue
-                            flags_blob = b" ".join(
-                                part[0] for part in msg_data
-                                if isinstance(part, tuple) and part and isinstance(part[0], (bytes, bytearray))
-                            )
-                            is_unread = b"\\Seen" not in flags_blob
                             # Headers + body land in different tuples in the
                             # response — concatenate the bytes for parsing.
                             raw = b""
@@ -2043,7 +1635,6 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                                 "headers": header_blob,
                                 "body": body_snippet.strip(),
                                 "message_id": (msg.get("Message-ID") or "").strip(),
-                                "unread": is_unread,
                             })
                         except Exception as _fe:
                             logger.debug(f"urgency: header fetch for uid {uid} failed: {_fe}")
@@ -2061,33 +1652,25 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
             for item in items:
                 scanned += 1
                 key = item["key"]
-                if item.get("unread"):
-                    all_unread_keys.add(key)
+                all_unread_keys.add(key)
                 if item.get("cached"):
-                    cached_v = dict(item["cached"])
-                    cached_v["unread"] = bool(item.get("unread"))
-                    per_uid_scores[key] = cached_v
+                    per_uid_scores[key] = item["cached"]
                     continue
                 # Skip uids we couldn't fetch (no subject/from/body).
                 if not item.get("subject") and not item.get("from"):
                     continue
-                verdict = _heuristic_email_verdict(item)
-                cache.setdefault("uids", {})[item["uid"]] = verdict
-                per_uid_scores[key] = verdict
-                saved_classifications += 1
-                continue
                 # ── LLM-classify. JSON-only response; bullet-proof parse.
                 llm_attempts += 1
                 prompt = (
-                    "You are triaging ONE email. Return ONLY JSON: "
+                    "You are triaging ONE unread email. Return ONLY JSON: "
                     "{\"score\":0|1|2|3,\"tags\":[\"...\"],\"spam\":false,"
                     "\"reason\":\"one short phrase\"}.\n"
                     "0 = trivial / promotional · 1 = informational, no reply needed · "
                     "2 = should reply within a day · 3 = urgent, reply now (deadline, blocker).\n\n"
-                    "Allowed visible tags: urgent, reply-soon, action-needed, calendar, bills, receipt, travel.\n"
-                    "Use action-needed when the user likely needs to reply, pay, sign, book, or decide. "
-                    "Use bills for bills or debts, receipt for purchases/deliveries, travel for reservations/trips, "
-                    "and calendar only when a calendar event/reminder is involved. spam=true for scams, phishing, "
+                    "Allowed tags: newsletter, marketing, notification, finance, bills, receipt, "
+                    "travel, security, shopping, social, work, personal, calendar.\n"
+                    "Use marketing for ads, promos, sales, offers, and cold sales. Use newsletter "
+                    "for newsletters, digests, and recurring content. spam=true for scams, phishing, "
                     "junk, cold sales, generic ads, or no-personal-action bulk mail.\n"
                     "Important: 'I'm outside', 'I am outside', 'waiting outside', 'at the door', "
                     "'locked out', or 'can't get in' means score 3 unless clearly historical.\n\n"
@@ -2096,7 +1679,6 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                     f"Snippet:\n{item.get('body','')}\n"
                 )
                 try:
-                    await wait_for_interactive_quiet("email urgency action")
                     raw = await llm_call_async_with_fallback(
                         candidates,
                         [{"role": "user", "content": prompt}],
@@ -2157,10 +1739,14 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                         r"\b(advertisement|sponsored|promo|promotion|sale|discount|offer|limited time|deal|tickets?|tour|merch|stream|purchase|sold out|low tickets|coupon|shop now|buy now)\b",
                         _blob,
                     ))
+                    if "newsletter" not in tags and bulkish:
+                        tags.append("newsletter")
+                    if "marketing" not in tags and marketingish:
+                        tags.append("marketing")
                     if (bulkish or marketingish) and score < 2:
                         score = 0
                         if not reason or "urgent" in reason.lower():
-                            reason = "bulk mail; no personal reply needed"
+                            reason = "Bulk marketing/newsletter; no personal reply needed"
                     # Strip "Name <addr>" to bare display name for compact summary.
                     _from_raw = item.get("from", "") or ""
                     if "<" in _from_raw:
@@ -2178,7 +1764,6 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                         # Cache the message_id too so re-scans of already-cached
                         # UIDs can still write the inbox tag without re-LLM'ing.
                         "message_id": (item.get("message_id") or "").strip(),
-                        "unread": bool(item.get("unread")),
                         "ts": _time.time(),
                     }
                     cache.setdefault("uids", {})[item["uid"]] = verdict
@@ -2193,9 +1778,9 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                     logger.debug(f"urgency: LLM classify failed for {key}: {e}")
                     continue
 
-            # ── Prune cache entries for UIDs that are no longer in the recent
-            # scan window. Read messages remain cached because tags are useful
-            # on read mail too; unread state is refreshed per scan above.
+            # ── Prune cache entries for UIDs that are no longer unread (replied
+            # / archived / deleted). Compare against `items` (everything UNSEEN
+            # in this scan window).
             seen_uids = {it["uid"] for it in items}
             cache_uids = cache.get("uids", {})
             for stale in [u for u in cache_uids if u not in seen_uids]:
@@ -2230,17 +1815,15 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                         _tag = str(_tag).strip().lower().replace("_", "-")
                         if _tag == "promo":
                             _tag = "marketing"
-                        if _tag == "action-needed" and any(t in _new_tags for t in ("urgent", "reply-soon")):
-                            continue
-                        if _tag in VISIBLE_EMAIL_TAGS and _tag not in _new_tags:
+                        if _tag in CATEGORY_TAGS and _tag not in _new_tags:
                             _new_tags.append(_tag)
                     _spam = 1 if _v.get("spam") else 0
                     # _key is "<account_id>:<uid>" — extract uid for the row.
-                    _acc_id, _uid_only = (_key.split(":", 1) + [""])[:2]
+                    _uid_only = _key.split(":", 1)[-1]
                     _owner_key = owner or ""
                     _row = _conn.execute(
-                        "SELECT tags FROM email_tags WHERE message_id=? AND owner=? AND account_id=?",
-                        (_msg_id, _owner_key, _acc_id),
+                        "SELECT tags FROM email_tags WHERE message_id=? AND owner=?",
+                        (_msg_id, _owner_key),
                     ).fetchone()
                     if _row:
                         try:
@@ -2259,42 +1842,23 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
                         for _tag in _new_tags:
                             if _tag not in _existing:
                                 _existing.append(_tag)
-                        if _new_tags or _spam:
-                            tag_write_details.append({
-                                "uid": _uid_only,
-                                "subject": _v.get("subject", ""),
-                                "from": _v.get("from", ""),
-                                "tags": list(_new_tags),
-                                "spam": _spam,
-                                "reason": _v.get("reason", ""),
-                                "updated": True,
-                            })
                         _conn.execute(
                             "UPDATE email_tags SET tags=?, spam_verdict=?, spam_reason=?, uid=?, folder=?, subject=?, sender=? "
-                            "WHERE message_id=? AND owner=? AND account_id=?",
+                            "WHERE message_id=? AND owner=?",
                             (_json.dumps(_existing), _spam, _v.get("reason", ""), _uid_only, "INBOX",
-                             _v.get("subject", ""), _v.get("from", ""), _msg_id, _owner_key, _acc_id),
+                             _v.get("subject", ""), _v.get("from", ""), _msg_id, _owner_key),
                         )
                     else:
                         if not _new_tags and not _spam:
                             continue
                         _conn.execute(
                             "INSERT INTO email_tags "
-                            "(message_id, owner, account_id, uid, folder, subject, sender, tags, spam_verdict, spam_reason, created_at) "
-                            "VALUES (?, ?, ?, ?, 'INBOX', ?, ?, ?, ?, ?, ?)",
-                            (_msg_id, _owner_key, _acc_id, _uid_only, _v.get("subject", ""),
+                            "(message_id, owner, uid, folder, subject, sender, tags, spam_verdict, spam_reason, created_at) "
+                            "VALUES (?, ?, ?, 'INBOX', ?, ?, ?, ?, ?, ?)",
+                            (_msg_id, _owner_key, _uid_only, _v.get("subject", ""),
                              _v.get("from", ""), _json.dumps(_new_tags), _spam, _v.get("reason", ""),
                              _dt2.utcnow().isoformat()),
                         )
-                        tag_write_details.append({
-                            "uid": _uid_only,
-                            "subject": _v.get("subject", ""),
-                            "from": _v.get("from", ""),
-                            "tags": list(_new_tags),
-                            "spam": _spam,
-                            "reason": _v.get("reason", ""),
-                            "updated": False,
-                        })
                 _conn.commit()
             finally:
                 _conn.close()
@@ -2302,7 +1866,7 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
             logger.warning(f"urgency: bulk tag write failed: {_te}")
 
         # ── 4. Aggregate state. urgent = score ≥ 2.
-        urgent_keys = [k for k, v in per_uid_scores.items() if v.get("score", 0) >= 2 and v.get("unread")]
+        urgent_keys = [k for k, v in per_uid_scores.items() if v.get("score", 0) >= 2]
         max_score = max((v.get("score", 0) for v in per_uid_scores.values()), default=0)
         total_urgent = len(urgent_keys)
 
@@ -2411,28 +1975,13 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
             f"reply-soon {tier_counts[2]} · info {tier_counts[1]} · trivial {tier_counts[0]} · "
             f"{saved_classifications} saved classifications"
         )
-        if failed_classifications:
-            head += f" · {len(failed_classifications)} failed"
+        if llm_attempts != saved_classifications:
+            head += f" · {llm_attempts - saved_classifications} failed"
         if newly_notified:
             head += f" · notified {len(newly_notified)}"
         if notify_failed:
             head += f" · notify failed {len(notify_failed)}"
 
-        def _fmt_tag_write(v):
-            subj = (v.get("subject") or "(no subject)")[:80]
-            frm = v.get("from") or ""
-            tags = list(v.get("tags") or [])
-            if v.get("spam"):
-                tags.append("spam")
-            tag_txt = ", ".join(tags) if tags else "cleared managed tags"
-            why = v.get("reason") or ""
-            op = "updated" if v.get("updated") else "created"
-            line = f"- **{subj}**" + (f" — _{frm}_" if frm else "")
-            line += f" — `{tag_txt}` ({op})"
-            if why:
-                line += f" · {why}"
-            return line
-
         def _fmt_one(v, newly_notified_set, failed_set, key):
             subj = (v.get("subject") or "(no subject)")[:80]
             frm = v.get("from") or ""
@@ -2448,13 +1997,6 @@ async def action_check_email_urgency(owner: str, **kwargs) -> Tuple[str, bool]:
         for k, v in per_uid_scores.items():
             by_tier.setdefault(v.get("score", 0), []).append((k, v))
         lines = [head]
-        if tag_write_details:
-            lines.append("")
-            lines.append(f"**Applied tags ({len(tag_write_details)}):**")
-            for v in tag_write_details[:16]:
-                lines.append(_fmt_tag_write(v))
-            if len(tag_write_details) > 16:
-                lines.append(f"…and {len(tag_write_details) - 16} more")
         tier_labels = {3: "Urgent", 2: "Reply soon", 1: "Informational", 0: "Trivial"}
         for tier in (3, 2, 1, 0):
             items_t = by_tier.get(tier, [])
@@ -2526,7 +2068,6 @@ async def action_cookbook_serve(
         end_after_min = int(cfg.get("end_after_min") or 0)
     except Exception:
         end_after_min = 0
-    set_default = bool(cfg.get("set_default", True))
 
     state_path = Path(COOKBOOK_STATE_FILE)
     try:
@@ -2613,51 +2154,6 @@ async def action_cookbook_serve(
         return f"Launch rejected: {data.get('error') or data.get('detail') or 'unknown'}", False
 
     sid = data.get("session_id") or ""
-    endpoint_id = data.get("endpoint_id") or ""
-    # Scheduled serves are usually meant to become the active local model for
-    # chat/tools while their time window is open. Persist both endpoint and
-    # model so task/utility/default resolution does not keep routing to a stale
-    # API fallback. Allow explicit opt-out with {"set_default": false}.
-    if endpoint_id and set_default:
-        try:
-            selected_model = repo_id
-            try:
-                from core.database import SessionLocal as _SL, ModelEndpoint as _ME
-                _db = _SL()
-                try:
-                    _ep = _db.query(_ME).filter(_ME.id == endpoint_id).first()
-                    if _ep and _ep.cached_models:
-                        _models = json.loads(_ep.cached_models or "[]")
-                        if isinstance(_models, list) and _models:
-                            selected_model = str(_models[0])
-                finally:
-                    _db.close()
-            except Exception:
-                pass
-            from src.settings import load_settings as _load_settings, save_settings as _save_settings
-            _settings = _load_settings()
-            _settings["default_endpoint_id"] = endpoint_id
-            _settings["default_model"] = selected_model
-            # Keep background tasks aligned unless the user explicitly chose a
-            # separate task model.
-            if not (_settings.get("task_endpoint_id") or "").strip():
-                _settings["task_endpoint_id"] = endpoint_id
-                _settings["task_model"] = selected_model
-            if not (_settings.get("utility_endpoint_id") or "").strip():
-                _settings["utility_endpoint_id"] = endpoint_id
-                _settings["utility_model"] = selected_model
-            _save_settings(_settings)
-            if owner:
-                from routes.prefs_routes import _load_for_user, _save_for_user
-                _prefs = _load_for_user(owner)
-                _prefs["default_endpoint_id"] = endpoint_id
-                _prefs["default_model"] = selected_model
-                if not (_prefs.get("utility_endpoint_id") or "").strip():
-                    _prefs["utility_endpoint_id"] = endpoint_id
-                    _prefs["utility_model"] = selected_model
-                _save_for_user(owner, _prefs)
-        except Exception as e:
-            logger.warning(f"cookbook_serve: default endpoint update failed: {e}")
     # Register the new task in cookbook_state.json + stamp it with our
     # scheduler-owner markers. /api/model/serve spawns the tmux session
     # but leaves the state-write to the UI — when a scheduled action
@@ -2679,8 +2175,6 @@ async def action_cookbook_serve(
             )
             if existing is None:
                 display_name = repo_id.split("/")[-1] if "/" in repo_id else repo_id
-                ssh_port = str(srv.get("port") or cfg.get("ssh_port") or "")
-                platform = str(srv.get("platform") or cfg.get("platform") or "linux")
                 placeholder = (
                     f"Launched by scheduled task {task_name!r} — waiting for tmux output…\n"
                     f"  session: {sid}\n"
@@ -2698,19 +2192,15 @@ async def action_cookbook_serve(
                     "ts": int(_time.time() * 1000),
                     "payload": {"repo_id": repo_id, "remote_host": host or "", "_cmd": cmd},
                     "remoteHost": host or "",
-                    "sshPort": ssh_port or "",
-                    "platform": platform or "linux",
+                    "sshPort": "",
+                    "platform": "linux",
                     "_serveReady": False,
-                    "_endpointAdded": bool(endpoint_id),
+                    "_endpointAdded": False,
                 }
                 tasks.append(existing)
             # Stamp ownership + end-at on the task entry.
             existing["_scheduledByTask"] = task_name or ""
             existing["_scheduledByOwner"] = owner or ""
-            if endpoint_id:
-                existing["_endpointId"] = endpoint_id
-                existing["endpointId"] = endpoint_id
-                existing["_endpointAdded"] = True
             if end_after_min > 0:
                 existing["_scheduledStopAtMs"] = int(_time.time() * 1000) + end_after_min * 60 * 1000
             fresh["tasks"] = tasks
@@ -2738,7 +2228,6 @@ BUILTIN_ACTIONS = {
     "tidy_research": action_tidy_research,
     "summarize_emails": action_summarize_emails,
     "draft_email_replies": action_draft_email_replies,
-    "email_auto_translate": action_email_auto_translate,
     "extract_email_events": action_extract_email_events,
     "classify_events": action_classify_events,
     # ping_events removed from the user-facing registry. Calendar reminders
@@ -2763,7 +2252,6 @@ BUILTIN_ACTION_INFO = {
     "tidy_research": "Remove orphaned research files (sessions that were deleted)",
     "summarize_emails": "Pre-generate AI summaries for new inbox emails",
     "draft_email_replies": "Pre-draft AI reply suggestions for new inbox emails",
-    "email_auto_translate": "Detect foreign-language emails and cache translated text for the email reader",
     "extract_email_events": "Scan emails for booking/meeting confirmations and auto-add to calendar",
     "classify_events": "Tag upcoming events with importance (low/normal/high/critical) and type (work/health/travel/etc.); colors them too",
     "daily_brief": "Build a morning digest: today's calendar, unread email count + top senders, active todos",

src/builtin_mcp.py

@@ -89,21 +89,6 @@ _BUILTIN_NPX_SERVERS = {
 MCP_DISABLED = os.environ.get("ODYSSEUS_DISABLE_MCP", "").lower() in ("1", "true", "yes")
 
 
-# Strong references to the fire-and-forget startup tasks scheduled below.
-# asyncio only keeps weak references to tasks created via create_task, so
-# without this the GC can collect a task mid-execution and the server
-# registration silently never runs. Mirrors _spawn_bg in routes/chat_helpers.py.
-_BG_TASKS: set[asyncio.Task] = set()
-
-
-def _spawn_bg(coro) -> asyncio.Task:
-    """Schedule a background task and hold a strong reference until it finishes."""
-    task = asyncio.create_task(coro)
-    _BG_TASKS.add(task)
-    task.add_done_callback(_BG_TASKS.discard)
-    return task
-
-
 async def register_builtin_servers(mcp_manager):
     """Connect all built-in MCP servers to the manager."""
     if MCP_DISABLED:
@@ -138,7 +123,7 @@ async def register_builtin_servers(mcp_manager):
         if not os.path.exists(script_path):
             logger.warning(f"Built-in MCP server script not found: {script_path}")
             continue
-        _spawn_bg(_connect_python_server(server_id, script_path, name))
+        asyncio.create_task(_connect_python_server(server_id, script_path, name))
 
     # Register NPX-based servers in the background (they take longer to start)
     npx_path = _find_npx()
@@ -190,7 +175,7 @@ async def register_builtin_servers(mcp_manager):
             except BaseException as e:
                 logger.warning(f"Built-in NPX server {cfg['name']} error: {type(e).__name__}: {e}")
 
-    _spawn_bg(_start_npx_servers())
+    asyncio.create_task(_start_npx_servers())
 
 
 def _npx_package_from_args(args):
@@ -248,15 +233,6 @@ async def _is_npx_package_cached(npx_path, package_spec, timeout_s=5):
         except Exception:
             pass
         return False
-    except asyncio.CancelledError:
-        # The probe was cancelled (e.g. app shutdown). Reap the child so it
-        # isn't orphaned, then propagate the cancellation.
-        try:
-            proc.kill()
-            await proc.wait()
-        except Exception:
-            pass
-        raise
     return proc.returncode == 0 and bool(stdout.strip())
 
 

src/caldav_sync.py

@@ -25,7 +25,6 @@ Design notes:
 import asyncio
 import hashlib
 import ipaddress
-import json
 import logging
 import os
 import socket
@@ -275,7 +274,6 @@ def _sync_blocking(owner: str, url: str, username: str, password: str, account_i
     # the integrations form still works, sync just no-ops with an error.
     from caldav.lib.error import AuthorizationError, NotFoundError
     from core.database import CalendarCal, CalendarEvent, SessionLocal
-    from routes.calendar_routes import _ensure_positive_duration
 
     result = {"calendars": 0, "events": 0, "deleted": 0, "errors": []}
 
@@ -392,11 +390,6 @@ def _sync_blocking(owner: str, url: str, username: str, password: str, account_i
                             end_dt = start_dt + timedelta(days=1)
                         else:
                             end_dt = start_dt + timedelta(hours=1)
-                        # A synced event with DTEND <= DTSTART (e.g. a single-day
-                        # all-day event whose source wrote DTEND equal to DTSTART)
-                        # would be stored zero-duration and silently dropped by the
-                        # list_events overlap filter. Clamp to a positive span.
-                        end_dt = _ensure_positive_duration(start_dt, end_dt, all_day)
 
                         # is_utc reflects whether the source carried a TZ
                         # we converted from. All-day = no TZ semantics.
@@ -501,7 +494,6 @@ def _event_payload(ev) -> dict:
         "all_day": ev.all_day,
         "is_utc": ev.is_utc,
         "rrule": ev.rrule or "",
-        "recurrence_exdates": json.loads(ev.recurrence_exdates or "[]") if getattr(ev, "recurrence_exdates", "") else [],
     }
 
 

src/caldav_writeback.py

@@ -33,8 +33,7 @@ def build_event_ical(ev: dict) -> str:
     """Serialize a local event dict to a VCALENDAR/VEVENT iCalendar string.
 
     ``ev`` keys: uid, summary, description, location, dtstart (datetime),
-    dtend (datetime), all_day (bool), is_utc (bool), rrule (str),
-    recurrence_exdates (list[str]).
+    dtend (datetime), all_day (bool), is_utc (bool), rrule (str).
     Mirrors how the pull path interprets is_utc/all_day so a round-trip is stable.
     """
     from icalendar import Calendar, Event as iEvent
@@ -71,15 +70,6 @@ def build_event_ical(ev: dict) -> str:
             ve.add("rrule", vRecur.from_ical(ev["rrule"]))
         except Exception:
             logger.debug("CalDAV write-back: skipping unparseable rrule %r", ev.get("rrule"))
-    for exdate in ev.get("recurrence_exdates") or []:
-        try:
-            if ev.get("all_day"):
-                ve.add("exdate", datetime.strptime(exdate[:10], "%Y-%m-%d").date())
-            else:
-                dt = datetime.strptime(exdate[:16], "%Y-%m-%dT%H:%M")
-                ve.add("exdate", dt.replace(tzinfo=timezone.utc) if ev.get("is_utc") else dt)
-        except Exception:
-            logger.debug("CalDAV write-back: skipping unparseable exdate %r", exdate)
 
     cal.add_component(ve)
     return cal.to_ical().decode("utf-8")

src/chat_handler.py

@@ -29,34 +29,6 @@ from src.youtube_handler import (
 logger = logging.getLogger(__name__)
 
 
-def _sync_upload_vision_to_gallery(file_info: Dict[str, Any], owner: Optional[str], text: str) -> None:
-    file_hash = (file_info or {}).get("hash")
-    if not file_hash or not text:
-        return
-    try:
-        from core.database import GalleryImage, SessionLocal
-        db = SessionLocal()
-        try:
-            q = db.query(GalleryImage).filter(
-                GalleryImage.file_hash == file_hash,
-                GalleryImage.is_active == True,  # noqa: E712
-            )
-            if owner:
-                q = q.filter(GalleryImage.owner == owner)
-            img = q.first()
-            if not img:
-                return
-            img.caption = text.strip()
-            db.commit()
-        except Exception:
-            db.rollback()
-            raise
-        finally:
-            db.close()
-    except Exception as e:
-        logger.warning("Failed to sync upload vision text to gallery: %s", e)
-
-
 class ChatHandler:
     """Handles chat operations for both streaming and non-streaming endpoints."""
 
@@ -235,7 +207,6 @@ class ChatHandler:
                                     _vtext = _vf.read().strip()
                                 if _vtext:
                                     enhanced_message += f"\n[User-corrected caption / OCR for this image — treat as authoritative]:\n{_vtext}"
-                                    _sync_upload_vision_to_gallery(file_info, owner, _vtext)
                                     _m = meta_by_id.get(att_id)
                                     if _m is not None:
                                         _m["vision"] = _vtext
@@ -255,7 +226,6 @@ class ChatHandler:
                                     cached_desc = _vf.read().strip()
                                 if cached_desc and not cached_desc.startswith("["):
                                     vl_desc = cached_desc
-                                    _sync_upload_vision_to_gallery(file_info, owner, vl_desc)
                             except Exception:
                                 vl_desc = None
                         if not vl_desc:
@@ -267,7 +237,6 @@ class ChatHandler:
                                     os.makedirs(os.path.join(UPLOAD_DIR, ".vision"), exist_ok=True)
                                     with open(_vcache, "w", encoding="utf-8") as _vf:
                                         _vf.write(vl_desc)
-                                    _sync_upload_vision_to_gallery(file_info, owner, vl_desc)
                                 except Exception:
                                     pass
                         enhanced_message = f"{enhanced_message}\n\n[Image: {file_info['name']}]\n{vl_desc}"

src/chat_processor.py

@@ -12,45 +12,6 @@ from src.prompt_security import UNTRUSTED_CONTEXT_POLICY, untrusted_context_mess
 
 logger = logging.getLogger(__name__)
 
-
-def _clean_search_query(query: str, max_len: int = 200) -> str:
-    """Strip fenced code blocks from a search query while preserving inline
-    code text.
-
-    This is a focused, defensive cleanup for the *final* web-search query
-    selected in ``build_context_preface`` (issue #4547): regardless of whether
-    the query came from the LLM-generated path (#4557) or the first-line
-    fallback, residual fenced / inline markdown should not leak into the search
-    call. Rather than using regex (which is brittle and strips inline code
-    text like ``git reset`` from the query), we render the query to HTML via
-    ``markdown`` and parse it with ``BeautifulSoup`` so that:
-
-    * ``<pre>`` blocks (fenced / indented code) are removed entirely.
-    * ``<code>`` elements (inline code) are preserved as plain text.
-
-    Both libraries are already project dependencies. The result is whitespace
-    collapsed and truncated to ``max_len``; an all-code input collapses to an
-    empty string, which the caller treats as "no query".
-    """
-    import markdown as _md
-    from bs4 import BeautifulSoup as _BS
-
-    html = _md.markdown(query, extensions=["fenced_code"])
-    soup = _BS(html, "html.parser")
-
-    # Remove fenced / indented code blocks.
-    for pre in soup.find_all("pre"):
-        pre.decompose()
-
-    # Preserve inline code by unwrapping <code> to text.
-    for code in soup.find_all("code"):
-        code.replace_with(code.get_text())
-
-    text = soup.get_text(" ", strip=True)
-    text = re.sub(r"\s+", " ", text)
-    return text[:max_len]
-
-
 # ── Stopwords & tokenizer ──
 
 _STOPWORDS = frozenset(
@@ -319,61 +280,10 @@ class ChatProcessor:
         web_sources = []
         if use_web:
             try:
-                from src.llm_core import llm_call
-
-                t_url, t_model, t_headers = session.endpoint_url, session.model, session.headers
-
-                # Default fallback is the first non-empty line of the original user message
-                fallback_query = next((line.strip() for line in message.split("\n") if line.strip()), "")
-                search_query = fallback_query
-
-                try:
-                    generated_query = llm_call(
-                        t_url,
-                        t_model,
-                        [
-                            {
-                                "role": "system",
-                                "content": (
-                                    "Extract a concise search query from the user's message. "
-                                    "Reply ONLY with the query."
-                                ),
-                            },
-                            {"role": "user", "content": message},
-                        ],
-                        headers=t_headers,
-                        temperature=0.1,
-                        max_tokens=50,
-                        timeout=15,
-                    ).strip()
-
-                    if generated_query:
-                        # LLM successfully generated a non-empty query -> use the generated query
-                        search_query = generated_query
-                    else:
-                        # LLM returned an empty or whitespace-only query -> fall back to original query
-                        logger.warning("LLM generated an empty search query, using fallback.")
-                except Exception as e:
-                    # LLM failed (exception/error) -> fall back to original user query
-                    logger.warning(f"Failed to generate search query via LLM, using fallback: {e}")
-
-                search_query = " ".join(search_query.split())
-                if len(search_query) > 150:
-                    search_query = search_query[:150].strip()
-
-                # Defensive cleanup of the final selected query (interim fix
-                # for #4547): strip any residual fenced/inline markdown so that
-                # neither the generated query nor the first-line fallback leaks
-                # fences or backticks into the search call. No-op on clean
-                # generated queries; collapses to "" when the query is all code.
-                search_query = _clean_search_query(search_query, max_len=150)
-
-                if search_query:
-                    # Execute web search using the final selected query
-                    web_context, web_sources = comprehensive_web_search(
-                        search_query, time_filter=time_filter, return_sources=True
-                    )
-                    preface.append(untrusted_context_message("web search results", web_context))
+                web_context, web_sources = comprehensive_web_search(
+                    message, time_filter=time_filter, return_sources=True
+                )
+                preface.append(untrusted_context_message("web search results", web_context))
             except Exception as e:
                 logger.error(f"Web search failed: {e}")
                 preface.append({"role": "system", "content": "Web search encountered an error and could not retrieve results."})

src/constants.py

@@ -4,7 +4,7 @@ import os
 
 from src.runtime_paths import get_app_root, get_default_data_dir
 
-APP_VERSION = "1.0.2"
+APP_VERSION = "1.0.1"
 
 # Base paths
 BASE_DIR = os.path.join(get_app_root(), "")

src/context_budget.py

@@ -18,13 +18,6 @@ DEFAULT_BUDGET = 6000
 DEFAULT_HEADROOM = 0.85
 
 
-def _int_or_zero(value) -> int:
-    try:
-        return int(value or 0)
-    except (TypeError, ValueError):
-        return 0
-
-
 def compute_input_token_budget(
     configured: int,
     context_length: int,
@@ -55,8 +48,8 @@ def compute_input_token_budget(
         - When the window is unknown (context_length <= 0), use the conservative
           ``default`` budget and do NOT scale off the fallback.
     """
-    configured = _int_or_zero(configured)
-    context_length = _int_or_zero(context_length)
+    configured = int(configured or 0)
+    context_length = int(context_length or 0)
 
     if explicit and configured > 0:
         return min(configured, context_length) if context_length > 0 else configured

src/cookbook_serve_lifecycle.py

@@ -37,13 +37,6 @@ async def _delete_endpoint_for_task(task: dict) -> None:
     the picker (probe goes offline; chats still try to route there) and
     the user has to delete it by hand in Settings -> Endpoints.
     """
-    endpoint_id = (task.get("_endpointId") or task.get("endpointId") or "").strip()
-    if not endpoint_id:
-        logger.info(
-            "cookbook_serve_lifecycle: task %s has no endpoint id; skipping endpoint deletion",
-            task.get("sessionId") or task.get("id") or "",
-        )
-        return
     import re as _re
     payload = task.get("payload") or {}
     cmd = str(payload.get("_cmd") or "")
@@ -73,10 +66,13 @@ async def _delete_endpoint_for_task(task: dict) -> None:
             if r.status_code >= 400:
                 return
             eps = r.json() if r.content else []
-            # Delete only the endpoint created by this scheduled serve. URL
-            # matching is unsafe because a later scheduled serve can reuse the
-            # same host:port after an older task has gone stale.
-            ep = next((e for e in eps if e.get("id") == endpoint_id), None)
+            # Prefer exact URL match; fall back to host:port substring so we
+            # still catch the case where 0.0.0.0 vs the registered host
+            # representation diverged.
+            ep = next((e for e in eps if e.get("base_url") == base_url), None)
+            if not ep:
+                hostport = f"{host}:{port}"
+                ep = next((e for e in eps if hostport in (e.get("base_url") or "")), None)
             if ep:
                 await client.delete(
                     f"{internal_api_base()}/api/model-endpoints/{ep['id']}",

src/document_actions.py

@@ -6,7 +6,7 @@ Reusable document actions callable from both REST routes and the task scheduler.
 
 import logging
 import re
-from datetime import datetime, timezone
+from datetime import datetime
 
 logger = logging.getLogger(__name__)
 
@@ -77,28 +77,10 @@ async def run_document_tidy(owner: str) -> str:
         deleted = 0
         kept = 0
         survivors = []  # docs that pass the junk rules, considered for dedup
-        now = datetime.now(timezone.utc)
 
         for doc in docs:
-            created = doc.created_at
-            if created and created.tzinfo is None:
-                created = created.replace(tzinfo=timezone.utc)
-
-            # Skip freshly created documents to avoid deleting them while the user is actively editing
-            if created and (now - created).total_seconds() < 900:  # 15 minutes
-                survivors.append(doc)
-                continue
-
             content = (doc.current_content or "").strip()
             title = (doc.title or "").strip().lower()
-            is_fresh_empty = (
-                not content
-                and created is not None
-                and (now - created).total_seconds() < 1800
-            )
-            if is_fresh_empty:
-                survivors.append(doc)
-                continue
 
             # Strip markdown noise to get "real" character count
             stripped = re.sub(r"^#{1,6}\s+", "", content, flags=re.MULTILINE)  # headers

src/embeddings.py

@@ -55,8 +55,6 @@ class EmbeddingClient:
         # of stalling startup ~30s per probe. Read stays generous for a real
         # endpoint (embedding a short string returns in well under a second).
         self._client = httpx.Client(timeout=httpx.Timeout(connect=3.0, read=10.0, write=5.0, pool=3.0))
-        self._batch_size = max(1, int(os.getenv("EMBEDDING_BATCH_SIZE", "8")))
-        self._max_chars = max(200, int(os.getenv("EMBEDDING_MAX_CHARS", "900")))
 
     def get_sentence_embedding_dimension(self) -> int:
         """Probe the endpoint for embedding dimension if not yet known."""
@@ -75,10 +73,23 @@ class EmbeddingClient:
         if not texts:
             return np.array([], dtype="float32")
 
+        # Batch in chunks of 64 to avoid oversized requests
         all_vecs = []
-        for i in range(0, len(texts), self._batch_size):
-            batch = texts[i : i + self._batch_size]
-            all_vecs.extend(self._embed_batch(batch))
+        for i in range(0, len(texts), 64):
+            batch = texts[i : i + 64]
+            resp = self._client.post(
+                self.url,
+                headers={"Authorization": f"Bearer {self.api_key}"} if self.api_key else {},
+                json={"input": batch, "model": self.model},
+            )
+            resp.raise_for_status()
+            data = resp.json()
+
+            # OpenAI format: {"data": [{"embedding": [...], "index": 0}, ...]}
+            embeddings = data.get("data", [])
+            embeddings.sort(key=lambda e: e.get("index", 0))
+            for emb in embeddings:
+                all_vecs.append(emb["embedding"])
 
         vecs = np.array(all_vecs, dtype="float32")
 
@@ -92,42 +103,6 @@ class EmbeddingClient:
 
         return vecs
 
-    def _embed_batch(self, batch: List[str]) -> List[List[float]]:
-        try:
-            return self._post_embeddings(batch)
-        except httpx.HTTPStatusError as e:
-            status = e.response.status_code if e.response is not None else None
-            if status != 400:
-                raise
-            if len(batch) > 1:
-                vecs = []
-                for text in batch:
-                    vecs.extend(self._embed_batch([text]))
-                return vecs
-            text = batch[0]
-            trimmed = text[: self._max_chars]
-            if trimmed != text:
-                logger.warning(
-                    "Embedding input exceeded endpoint context; retrying with %d chars",
-                    len(trimmed),
-                )
-                return self._post_embeddings([trimmed])
-            raise
-
-    def _post_embeddings(self, batch: List[str]) -> List[List[float]]:
-        resp = self._client.post(
-            self.url,
-            headers={"Authorization": f"Bearer {self.api_key}"} if self.api_key else {},
-            json={"input": batch, "model": self.model},
-        )
-        resp.raise_for_status()
-        data = resp.json()
-
-        # OpenAI format: {"data": [{"embedding": [...], "index": 0}, ...]}
-        embeddings = data.get("data", [])
-        embeddings.sort(key=lambda e: e.get("index", 0))
-        return [emb["embedding"] for emb in embeddings]
-
 
 class FastEmbedClient:
     """Local embedding client using fastembed (ONNX). No external service needed."""

src/endpoint_resolver.py

@@ -47,33 +47,6 @@ def _endpoint_cached_models(ep) -> list:
     return models if isinstance(models, list) else []
 
 
-def _endpoint_pinned_models(ep) -> list:
-    raw = getattr(ep, "pinned_models", None)
-    if not raw:
-        return []
-    try:
-        models = json.loads(raw) if isinstance(raw, str) else raw
-    except Exception:
-        return []
-    return models if isinstance(models, list) else []
-
-
-def _is_mlx_deepseek_v4_repo_id(model_id: str) -> bool:
-    return "mlx-community/deepseek-v4" in str(model_id or "").lower()
-
-
-def _is_mlx_deepseek_v4_shim_id(model_id: str) -> bool:
-    return "/.cache/odysseus/mlx-shims/deepseek-v4" in str(model_id or "").lower()
-
-
-def _filter_mlx_deepseek_v4_repo_when_shimmed(model_ids) -> list:
-    ids = list(model_ids or [])
-    has_shim = any(_is_mlx_deepseek_v4_shim_id(m) for m in ids)
-    if not has_shim:
-        return ids
-    return [m for m in ids if not _is_mlx_deepseek_v4_repo_id(m)]
-
-
 def _endpoint_hidden_models(ep) -> set:
     """Model ids the admin disabled on this endpoint (the UI's hidden list)."""
     raw = getattr(ep, "hidden_models", None)
@@ -94,15 +67,7 @@ def _endpoint_enabled_models(ep) -> list:
     raw first one resolves to a model that 400s ("requires terms acceptance").
     """
     hidden = _endpoint_hidden_models(ep)
-    merged = []
-    seen = set()
-    for m in [*_endpoint_cached_models(ep), *_endpoint_pinned_models(ep)]:
-        if not isinstance(m, str) or not m or m in seen:
-            continue
-        seen.add(m)
-        merged.append(m)
-    merged = _filter_mlx_deepseek_v4_repo_when_shimmed(merged)
-    return [m for m in merged if m not in hidden]
+    return [m for m in _endpoint_cached_models(ep) if m not in hidden]
 
 
 def resolve_endpoint_runtime(ep, owner: Optional[str] = None) -> Tuple[str, Optional[str]]:

src/host_docker_access.py

@@ -1,62 +0,0 @@
-"""Policy checks for explicit host Docker access from a container."""
-
-import os
-import stat
-from collections.abc import Mapping
-
-
-HOST_DOCKER_ENV_VAR = "ODYSSEUS_ENABLE_HOST_DOCKER"
-HOST_DOCKER_SOCKET_PATH = "/var/run/docker.sock"
-
-HOST_DOCKER_ACCESS_HINT = (
-    "Local Docker daemon access is disabled inside the Odysseus container; a "
-    "Docker CLI alone is not enough. Default Docker Compose intentionally does "
-    "not mount the host Docker socket. Raw socket access is high-trust and can "
-    "grant broad control over the host Docker daemon. If you accept that risk, "
-    "enable docker/host-docker.yml. Remote server Docker workflows over SSH "
-    "remain preferred."
-)
-
-
-def running_in_container(
-    dockerenv_path: str = "/.dockerenv",
-    cgroup_path: str = "/proc/1/cgroup",
-) -> bool:
-    if os.path.exists(dockerenv_path):
-        return True
-    try:
-        with open(cgroup_path, "r", encoding="utf-8") as handle:
-            contents = handle.read()
-    except OSError:
-        return False
-    return any(token in contents for token in ("docker", "containerd", "kubepods"))
-
-
-def host_docker_access_enabled(
-    socket_path: str = HOST_DOCKER_SOCKET_PATH,
-    *,
-    environ: Mapping[str, str] | None = None,
-) -> bool:
-    env = os.environ if environ is None else environ
-    if env.get(HOST_DOCKER_ENV_VAR, "").strip().lower() != "true":
-        return False
-    try:
-        mode = os.stat(socket_path).st_mode
-    except OSError:
-        return False
-    return stat.S_ISSOCK(mode)
-
-
-def local_docker_available(
-    *,
-    cli_available: bool,
-    in_container: bool | None = None,
-    environ: Mapping[str, str] | None = None,
-    socket_path: str = HOST_DOCKER_SOCKET_PATH,
-) -> bool:
-    if not cli_available:
-        return False
-    containerized = running_in_container() if in_container is None else in_container
-    if not containerized:
-        return True
-    return host_docker_access_enabled(socket_path, environ=environ)

src/integrations.py

@@ -216,14 +216,7 @@ def _normalize_integration_base_url(base_url: Any) -> str:
 
 
 def _join_integration_url(base_url: str, path: str) -> str:
-    base = base_url.rstrip("/")
-    rel = path.lstrip("/")
-    if not rel:
-        # A bare "/" must resolve to the base URL itself, not base + "/".
-        # POST-to-base integrations (e.g. Discord webhooks) 404 on the
-        # trailing-slash variant of their URL.
-        return base
-    return urljoin(base + "/", rel)
+    return urljoin(base_url.rstrip("/") + "/", path.lstrip("/"))
 
 
 def load_integrations() -> List[Dict[str, Any]]:
@@ -401,22 +394,6 @@ async def execute_api_call(
         return {"error": "Path must not contain a fragment", "exit_code": 1}
 
     url = _join_integration_url(base_url, path)
-
-    # SSRF guard — same check used by the gallery endpoint, embeddings,
-    # CardDAV, and the reminder webhook sender. Link-local / metadata
-    # addresses (169.254.x.x — the cloud credential-exfil vector) are always
-    # rejected; INTEGRATION_API_BLOCK_PRIVATE_IPS=true also blocks RFC-1918 /
-    # loopback for locked-down deployments. Private stays allowed by default
-    # because LAN integrations (Home Assistant, Miniflux, ntfy) are the
-    # primary use case.
-    from src.url_safety import check_outbound_url
-    block_private = os.getenv(
-        "INTEGRATION_API_BLOCK_PRIVATE_IPS", "false"
-    ).lower() == "true"
-    ok, reason = check_outbound_url(url, block_private=block_private)
-    if not ok:
-        return {"error": f"URL rejected: {reason}", "exit_code": 1}
-
     method = method.upper()
 
     # Build headers

src/interactive_gate.py

@@ -1,203 +0,0 @@
-"""Foreground activity gate for background work.
-
-Background tasks are allowed to run only after normal UI/API traffic has
-settled. This keeps scheduled jobs and email pollers from competing with the
-user opening Odysseus, Cookbook, email, documents, notes, or other panels.
-"""
-
-from __future__ import annotations
-
-import asyncio
-from contextlib import asynccontextmanager
-import os
-import time
-
-
-_ACTIVE_REQUESTS = 0
-_LAST_ACTIVITY = 0.0
-_LAST_BROWSER_ACTIVITY = 0.0
-_COND: asyncio.Condition | None = None
-_COND_LOOP: asyncio.AbstractEventLoop | None = None
-
-
-def _enabled() -> bool:
-    return os.getenv("BACKGROUND_TASK_FOREGROUND_GATE", "true").lower() not in {"0", "false", "no", "off"}
-
-
-def _quiet_seconds() -> float:
-    try:
-        return max(0.0, float(os.getenv("BACKGROUND_TASK_QUIET_MS", "1500")) / 1000.0)
-    except Exception:
-        return 1.5
-
-
-def _max_wait_seconds() -> float:
-    """0 means wait indefinitely until the UI is quiet."""
-    try:
-        return max(0.0, float(os.getenv("BACKGROUND_TASK_MAX_WAIT_SECONDS", "0")))
-    except Exception:
-        return 0.0
-
-
-def _browser_active_seconds() -> float:
-    """How long a visible Odysseus browser heartbeat blocks background tasks."""
-    try:
-        return max(0.0, float(os.getenv("BACKGROUND_TASK_BROWSER_ACTIVE_SECONDS", "45")))
-    except Exception:
-        return 45.0
-
-
-def _condition() -> asyncio.Condition:
-    global _COND, _COND_LOOP
-    try:
-        loop = asyncio.get_running_loop()
-    except RuntimeError:
-        loop = None
-    if _COND is None or _COND_LOOP is not loop:
-        _COND = asyncio.Condition()
-        _COND_LOOP = loop
-    return _COND
-
-
-_PASSIVE_EXACT_PATHS = {
-    "/api/activity/heartbeat",
-    "/api/client-perf",
-    "/api/tasks/notifications",
-    "/api/research/active",
-    "/api/email/urgency-state",
-}
-
-_PASSIVE_PREFIXES = (
-    "/api/chat/stream_status",
-    "/api/health",
-    "/api/prefs",
-)
-
-
-def should_track_interactive_request(path: str, method: str = "GET") -> bool:
-    if not _enabled():
-        return False
-    if (method or "").upper() == "OPTIONS":
-        return False
-    if path in _PASSIVE_EXACT_PATHS:
-        return False
-    if any(path.startswith(prefix) for prefix in _PASSIVE_PREFIXES):
-        return False
-    return True
-
-
-async def mark_browser_activity() -> None:
-    """Record that an authenticated browser tab is visibly using Odysseus."""
-    global _LAST_BROWSER_ACTIVITY
-    if not _enabled():
-        return
-    cond = _condition()
-    async with cond:
-        _LAST_BROWSER_ACTIVITY = time.monotonic()
-        cond.notify_all()
-
-
-def _has_recent_browser_activity(now: float | None = None) -> bool:
-    ttl = _browser_active_seconds()
-    if ttl <= 0 or _LAST_BROWSER_ACTIVITY <= 0:
-        return False
-    return ((now if now is not None else time.monotonic()) - _LAST_BROWSER_ACTIVITY) < ttl
-
-
-def has_foreground_activity(now: float | None = None) -> bool:
-    """Return True when foreground browser/model work should stop background jobs.
-
-    Passive polling endpoints are excluded by should_track_interactive_request,
-    so active/recent request tracking is safe to use here. This matters during
-    initial page load: the heartbeat may not have landed yet, but the user is
-    already waiting on real UI requests.
-    """
-    if not _enabled():
-        return False
-    t = now if now is not None else time.monotonic()
-    if _ACTIVE_REQUESTS > 0:
-        return True
-    if _LAST_ACTIVITY > 0 and (t - _LAST_ACTIVITY) < _quiet_seconds():
-        return True
-    return _has_recent_browser_activity(t) or _has_active_chat_stream()
-
-
-def _has_active_chat_stream() -> bool:
-    """Best-effort check for foreground model work that outlives HTTP requests.
-
-    Chat/agent streams are detached from the browser SSE so a stream can keep
-    running after the request that started it has returned. Background LLM
-    tasks must still wait for those runs; otherwise helpers like email
-    auto-translate compete with the user's active chat on the same local model.
-    """
-    try:
-        from routes import chat_routes as _chat_routes
-        active_streams = getattr(_chat_routes, "_active_streams", {}) or {}
-        if active_streams:
-            return True
-    except Exception:
-        pass
-    try:
-        from src import agent_runs
-        runs = getattr(agent_runs, "_RUNS", {}) or {}
-        return any(getattr(run, "status", None) == "running" for run in runs.values())
-    except Exception:
-        return False
-
-
-@asynccontextmanager
-async def track_interactive_request(path: str = "", method: str = ""):
-    global _ACTIVE_REQUESTS, _LAST_ACTIVITY
-    if not _enabled():
-        yield
-        return
-
-    cond = _condition()
-    async with cond:
-        _ACTIVE_REQUESTS += 1
-        _LAST_ACTIVITY = time.monotonic()
-        cond.notify_all()
-    try:
-        yield
-    finally:
-        async with cond:
-            _ACTIVE_REQUESTS = max(0, _ACTIVE_REQUESTS - 1)
-            _LAST_ACTIVITY = time.monotonic()
-            cond.notify_all()
-
-
-async def wait_for_interactive_quiet(label: str = "") -> bool:
-    """Wait until foreground requests have stopped for the configured window.
-
-    Returns True if the caller had to wait at all. The label is intentionally
-    only for future logging/debugging so callers can keep their code simple.
-    """
-    if not _enabled():
-        return False
-
-    quiet = _quiet_seconds()
-    max_wait = _max_wait_seconds()
-    deadline = time.monotonic() + max_wait if max_wait > 0 else None
-    cond = _condition()
-    waited = False
-
-    while True:
-        async with cond:
-            now = time.monotonic()
-            quiet_remaining = quiet - (now - _LAST_ACTIVITY)
-            active_stream = _has_active_chat_stream()
-            browser_active = _has_recent_browser_activity(now)
-            if _ACTIVE_REQUESTS <= 0 and quiet_remaining <= 0 and not active_stream and not browser_active:
-                return waited
-
-            waited = True
-            timeout = 0.25 if (_ACTIVE_REQUESTS > 0 or active_stream or browser_active) else min(max(quiet_remaining, 0.05), 0.5)
-            if deadline is not None:
-                remaining = deadline - now
-                if remaining <= 0:
-                    return waited
-                timeout = min(timeout, remaining)
-            try:
-                await asyncio.wait_for(cond.wait(), timeout=timeout)
-            except asyncio.TimeoutError:
-                pass

src/llm_core.py

@@ -8,88 +8,13 @@ import hashlib
 import threading
 import re
 import os
-from contextlib import asynccontextmanager
 from fastapi import HTTPException
 from typing import Optional, Dict, List, Tuple
-from src.model_context import get_context_length, DEFAULT_CONTEXT, is_local_endpoint
+from src.model_context import get_context_length, DEFAULT_CONTEXT
 from urllib.parse import urlparse
 
 logger = logging.getLogger(__name__)
 
-_LOCAL_MODEL_LOCK = asyncio.Lock()
-_LOCAL_MODEL_WAITING_FOREGROUND = 0
-_LOCAL_MODEL_CURRENT: Dict[str, object] = {}
-
-
-def _local_model_gate_enabled() -> bool:
-    return os.getenv("ODYSSEUS_LOCAL_MODEL_GATE", "true").lower() not in {"0", "false", "no", "off"}
-
-
-def _gate_workload(workload: Optional[str]) -> str:
-    return "background" if str(workload or "").lower() == "background" else "foreground"
-
-
-@asynccontextmanager
-async def _local_model_slot(target_url: str, model: str, workload: Optional[str] = None):
-    """Serialize local model traffic, with foreground chat taking priority.
-
-    Most local servers expose one GPU/CPU generation pipe even when their HTTP
-    API accepts multiple requests. Letting scheduled email/tasks and foreground
-    chat hit that pipe together creates the user-visible "streams crossed" and
-    "prompt waited behind a task" failure mode. Cloud providers are left alone.
-    """
-    if not _local_model_gate_enabled() or not is_local_endpoint(target_url):
-        yield
-        return
-
-    global _LOCAL_MODEL_WAITING_FOREGROUND
-    kind = _gate_workload(workload)
-    current_task = asyncio.current_task()
-    if kind == "foreground":
-        _LOCAL_MODEL_WAITING_FOREGROUND += 1
-        current = dict(_LOCAL_MODEL_CURRENT)
-        if current.get("workload") == "background":
-            task = current.get("task")
-            if isinstance(task, asyncio.Task) and not task.done():
-                logger.info(
-                    "[model-gate] cancelling background local model call for foreground request model=%s",
-                    model,
-                )
-                task.cancel()
-    else:
-        # Background work should not jump in while the browser/chat is active
-        # or while a foreground request is waiting to acquire the local model.
-        try:
-            from src.interactive_gate import has_foreground_activity
-        except Exception:
-            has_foreground_activity = lambda: False  # type: ignore
-        while _LOCAL_MODEL_WAITING_FOREGROUND > 0 or has_foreground_activity():
-            await asyncio.sleep(0.25)
-
-    acquired = False
-    try:
-        await _LOCAL_MODEL_LOCK.acquire()
-        acquired = True
-        if kind == "foreground":
-            _LOCAL_MODEL_WAITING_FOREGROUND = max(0, _LOCAL_MODEL_WAITING_FOREGROUND - 1)
-        _LOCAL_MODEL_CURRENT.clear()
-        _LOCAL_MODEL_CURRENT.update({
-            "task": current_task,
-            "workload": kind,
-            "url": target_url,
-            "model": model,
-            "started": time.time(),
-        })
-        yield
-    finally:
-        if kind == "foreground":
-            _LOCAL_MODEL_WAITING_FOREGROUND = max(0, _LOCAL_MODEL_WAITING_FOREGROUND - 1)
-        if acquired and _LOCAL_MODEL_LOCK.locked():
-            owner = _LOCAL_MODEL_CURRENT.get("task")
-            if owner is current_task:
-                _LOCAL_MODEL_CURRENT.clear()
-            _LOCAL_MODEL_LOCK.release()
-
 class LLMConfig:
     """Configuration constants for LLM operations."""
     DEFAULT_TIMEOUT = 30
@@ -185,18 +110,6 @@ _HARMONY_MARKERS = (
 )
 _HARMONY_MAX_MARKER_LEN = max(len(marker) for marker in _HARMONY_MARKERS)
 
-_VISIBLE_CHAT_TEMPLATE_ARTIFACT_RE = re.compile(
-    r"(?:\|end\|)+\|?assistan(?:t)?\|?"
-    r"|\|assistan(?:t)?\|"
-    r"|<\|im_start\|>\s*assistant"
-    r"|<\|im_end\|>",
-    re.IGNORECASE,
-)
-
-
-def _strip_visible_chat_template_artifacts(text: str) -> str:
-    return _VISIBLE_CHAT_TEMPLATE_ARTIFACT_RE.sub("", text or "")
-
 
 def _harmony_suffix_hold_len(text: str) -> int:
     """Return how many trailing chars could be the start of a harmony marker."""
@@ -274,75 +187,6 @@ def _stream_delta_event(text: str, *, thinking: bool = False) -> str:
         payload["thinking"] = True
     return f"data: {json.dumps(payload)}\n\n"
 
-
-_DEGENERATE_WORD_RE = re.compile(r"[A-Za-z0-9_\u0370-\u03ff\u0400-\u04ff]+")
-
-
-class _DegenerateStreamGuard:
-    """Detect local-model token collapse before it floods the UI.
-
-    Some self-hosted models fail by repeating one token forever ("Var Var Var",
-    "Summer Summer ..."). This is not a useful response and can burn context,
-    browser memory, and GPU time. Keep the guard conservative: only fire on long
-    same-token runs or a very dominant repeated token in the recent window.
-    """
-
-    def __init__(self, model: str):
-        self.model = model or "model"
-        self.last_token = ""
-        self.same_run = 0
-        self.recent_tokens: List[str] = []
-        self.total_chars = 0
-
-    def check(self, text: str) -> Optional[str]:
-        if not text:
-            return None
-        self.total_chars += len(text)
-        tokens = [t.lower() for t in _DEGENERATE_WORD_RE.findall(text) if len(t) >= 2]
-        if not tokens:
-            return None
-        for token in tokens:
-            if token == self.last_token:
-                self.same_run += 1
-            else:
-                self.last_token = token
-                self.same_run = 1
-            self.recent_tokens.append(token)
-        if len(self.recent_tokens) > 96:
-            self.recent_tokens = self.recent_tokens[-96:]
-
-        reason = None
-        if self.same_run >= 28 and self.total_chars >= 100:
-            reason = f"repeated '{self.last_token}' {self.same_run} times"
-        elif len(self.recent_tokens) >= 72:
-            top = max(set(self.recent_tokens), key=self.recent_tokens.count)
-            count = self.recent_tokens.count(top)
-            if count >= 60 and count / max(len(self.recent_tokens), 1) >= 0.78:
-                reason = f"repeated '{top}' {count}/{len(self.recent_tokens)} recent tokens"
-        if not reason and len(self.recent_tokens) >= 80:
-            # Phrase loops are common on some local quantized MLX/MoE models:
-            # "Also be a software developer mode?" repeated forever will not
-            # trip the single-token guard above, but it is still a wedged
-            # generation. Require many repeats of the same 4-gram so normal
-            # prose/list formatting is not interrupted.
-            grams = [tuple(self.recent_tokens[i:i + 4]) for i in range(0, len(self.recent_tokens) - 3)]
-            if grams:
-                top_gram = max(set(grams), key=grams.count)
-                gram_count = grams.count(top_gram)
-                if gram_count >= 10:
-                    reason = f"repeated phrase '{' '.join(top_gram)}' {gram_count} times"
-
-        if not reason:
-            return None
-
-        logger.warning("[degenerate-stream] aborting model=%s reason=%s", self.model, reason)
-        message = (
-            f"Stopped generation: {self.model} started repeating tokens "
-            f"({reason}). Try a different model or lower temperature."
-        )
-        return f'event: error\ndata: {json.dumps({"status": 502, "text": message, "error": message})}\n\n'
-
-
 def _model_activity_key(url: str, model: str) -> str:
     return f"{(url or '').strip()}|{(model or '').strip()}"
 
@@ -501,18 +345,6 @@ def _normalize_ollama_url(url: str) -> str:
     return base.rstrip("/") + "/chat"
 
 
-def _normalize_openai_chat_url(url: str) -> str:
-    """Ensure an OpenAI-compatible base URL points at /chat/completions."""
-    base = (url or "").strip().rstrip("/")
-    if not base:
-        return base
-    if base.endswith("/chat/completions") or base.endswith("/completions"):
-        return base
-    if base.endswith("/models"):
-        base = base[: -len("/models")].rstrip("/")
-    return base + "/chat/completions"
-
-
 def _ollama_normalize_messages(messages: List[Dict]) -> List[Dict]:
     """Adapt Odysseus' canonical OpenAI-style messages to native Ollama /api/chat.
 
@@ -845,8 +677,6 @@ def _detect_provider(url: str) -> str:
     from src.copilot import is_copilot_base
     if is_copilot_base(url):
         return "copilot"
-    if _host_match(url, "cerebras.ai"):
-        return "cerebras"
     if _host_match(url, "mistral.ai"):
         return "mistral"
     return "openai"
@@ -899,52 +729,6 @@ def _apply_local_cache_affinity(payload: Dict, url: str, session_id: Optional[st
     payload.setdefault("cache_prompt", True)
 
 
-def _is_local_minimax_mlx_request(url: str, model: str) -> bool:
-    """Local MLX MiniMax-family endpoints need conservative sampling defaults.
-
-    The OpenAI-compatible MLX server accepts repetition/frequency penalties.
-    Some large quantized MiniMax/MoE ports otherwise fall into visible reasoning
-    loops ("Also be...", "No.", etc.) even for trivial prompts.
-    """
-    if not model:
-        return False
-    m = model.lower()
-    if "minimax" not in m and "mini-max" not in m:
-        return False
-    try:
-        from src.model_context import is_local_endpoint
-        return is_local_endpoint(url)
-    except Exception:
-        return False
-
-
-def _apply_local_generation_stability(payload: Dict, url: str, model: str) -> None:
-    if not _is_local_minimax_mlx_request(url, model):
-        return
-    if "temperature" in payload:
-        try:
-            # MiniMax MLX quantized ports are very sensitive to chat/agent
-            # harness size. Character presets can ask for a warmer voice, but
-            # local MiniMax needs a final compatibility clamp or trivial
-            # prompts can fall into visible reasoning/repetition loops.
-            payload["temperature"] = min(float(payload.get("temperature") or 0.2), 0.2)
-        except (TypeError, ValueError):
-            payload["temperature"] = 0.2
-    payload.setdefault("top_p", 0.9)
-    payload.setdefault("top_k", 20)
-    payload.setdefault("repetition_penalty", 1.12)
-    payload.setdefault("repetition_context_size", 256)
-    payload.setdefault("frequency_penalty", 0.08)
-    payload.setdefault("frequency_context_size", 256)
-    payload.setdefault("presence_penalty", 0.02)
-    payload.setdefault("presence_context_size", 256)
-    payload.setdefault("stop", ["<|im_end|>", "<|endoftext|>", "</s>"])
-    # A max_tokens of 0 means "server default/unbounded" for many local
-    # endpoints. Keep simple chats from running forever when the model loops.
-    if not payload.get("max_tokens") and not payload.get("max_completion_tokens"):
-        payload["max_tokens"] = 2048
-
-
 def _provider_headers(provider: str, headers: Optional[Dict] = None) -> Dict[str, str]:
     h = {"Content-Type": "application/json"}
     if isinstance(headers, dict):
@@ -979,8 +763,6 @@ def _provider_label(url: str) -> str:
     if is_chatgpt_subscription_base(url): return "ChatGPT Subscription"
     from src.copilot import is_copilot_base
     if is_copilot_base(url): return "GitHub Copilot"
-    if _host_match(url, "cerebras.ai"):
-        return "cerebras"
     if _host_match(url, "mistral.ai"): return "Mistral"
     if _host_match(url, "deepseek.com"): return "DeepSeek"
     if _host_match(url, "nvidia.com"): return "NVIDIA"
@@ -1414,25 +1196,6 @@ def _as_content_blocks(content) -> List[Dict]:
     return []
 
 
-def _is_untrusted_context_content(content) -> bool:
-    if isinstance(content, str):
-        return (
-            content.startswith("UNTRUSTED SOURCE DATA\n")
-            or "<<<UNTRUSTED_SOURCE_DATA>>>" in content
-        )
-    if isinstance(content, list):
-        return any(
-            isinstance(block, dict)
-            and block.get("type") == "text"
-            and _is_untrusted_context_content(block.get("text") or "")
-            for block in content
-        )
-    return False
-
-
-_REFERENCE_CONTEXT_BOUNDARY = "Reference context received."
-
-
 def _sanitize_llm_messages(messages: List[Dict]) -> List[Dict]:
     """Strip Odysseus-only metadata before sending messages to providers.
 
@@ -1545,10 +1308,6 @@ def _sanitize_llm_messages(messages: List[Dict]) -> List[Dict]:
 
         last = merged[-1]
         if last.get("role") == "user" and item.get("role") == "user":
-            if _is_untrusted_context_content(last.get("content")):
-                merged.append({"role": "assistant", "content": _REFERENCE_CONTEXT_BOUNDARY})
-                merged.append(item)
-                continue
             last_copy = dict(last)
             lc = last_copy.get("content")
             ic = item.get("content")
@@ -1575,7 +1334,6 @@ def _sanitize_llm_messages(messages: List[Dict]) -> List[Dict]:
 
     return merged
 
-
 def _normalize_anthropic_url(url: str) -> str:
     """Ensure Anthropic URL points to /v1/messages."""
     url = url.rstrip("/")
@@ -1687,10 +1445,8 @@ def list_model_ids(
         r = httpx_get_kimi_aware(models_url, h, timeout=timeout)
         r.raise_for_status()
         data = r.json()
-        # Some OpenAI-compatible APIs (e.g. Together) return a bare list here.
-        items = data if isinstance(data, list) else (data.get("data") or [])
-        model_ids = [m.get("id") for m in items if isinstance(m, dict) and m.get("id")]
-        if not model_ids and isinstance(data, dict):
+        model_ids = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
+        if not model_ids:
             model_ids = [
                 m.get("name") or m.get("model")
                 for m in (data.get("models") or [])
@@ -1778,7 +1534,7 @@ def llm_call(url: str, model: str, messages: List[Dict], temperature: float = LL
             stream=False, num_ctx=get_context_length(url, model),
         )
     else:
-        target_url = _normalize_openai_chat_url(url)
+        target_url = url
         if provider == "copilot":
             from src.copilot import apply_request_headers
             apply_request_headers(h, messages_copy)
@@ -1792,7 +1548,6 @@ def llm_call(url: str, model: str, messages: List[Dict], temperature: float = LL
         if max_tokens and max_tokens > 0:
             tok_key = "max_completion_tokens" if _uses_max_completion_tokens(model) else "max_tokens"
             payload[tok_key] = max_tokens
-        _apply_local_generation_stability(payload, target_url, model)
         if provider == "mistral" and _supports_thinking(model):
             payload["reasoning_effort"] = _MISTRAL_REASONING_EFFORT
     try:
@@ -1902,7 +1657,6 @@ async def llm_call_async(
     max_retries: int = LLMConfig.MAX_RETRIES,
     prompt_type: Optional[str] = None,
     session_id: Optional[str] = None,
-    workload: str = "foreground",
 ) -> str:
     """Asynchronous LLM call using httpx with connection pooling, timeout, retry logic, and performance logging."""
     provider = _detect_provider(url)
@@ -1940,7 +1694,6 @@ async def llm_call_async(
             max_tokens=max_tokens,
             headers=headers,
             timeout=timeout,
-            workload=workload,
         ):
             event_is_error = False
             for line in str(chunk).splitlines():
@@ -1985,7 +1738,7 @@ async def llm_call_async(
             stream=False, num_ctx=get_context_length(url, model),
         )
     else:
-        target_url = _normalize_openai_chat_url(url)
+        target_url = url
         h = _provider_headers(provider, headers)
         if provider == "copilot":
             from src.copilot import apply_request_headers
@@ -2006,7 +1759,6 @@ async def llm_call_async(
         if provider == "mistral" and _supports_thinking(model):
             payload["reasoning_effort"] = _MISTRAL_REASONING_EFFORT
         _apply_local_cache_affinity(payload, url, session_id)
-        _apply_local_generation_stability(payload, target_url, model)
 
     if _is_host_dead(target_url):
         raise HTTPException(503, f"Upstream {_host_key(target_url)} marked unreachable (cooldown active)")
@@ -2017,10 +1769,9 @@ async def llm_call_async(
         attempt += 1
         start = time.time()
         try:
-            async with _local_model_slot(target_url, model, workload):
-                note_model_activity(target_url, model)
-                client = _get_http_client()
-                r = await httpx_post_kimi_aware_async(client, target_url, h, json=payload, timeout=call_timeout)
+            note_model_activity(target_url, model)
+            client = _get_http_client()
+            r = await httpx_post_kimi_aware_async(client, target_url, h, json=payload, timeout=call_timeout)
             duration = time.time() - start
             if not r.is_success:
                 friendly = _format_upstream_error(r.status_code, r.text, target_url)
@@ -2062,45 +1813,10 @@ async def llm_call_async(
                 raise HTTPException(502, f"POST {target_url} failed after {max_retries} attempts: {e}")
             await asyncio.sleep(LLMConfig.RETRY_DELAY)
 
-def _stream_target_url(url: str) -> str:
-    provider = _detect_provider(url)
-    if provider == "anthropic":
-        return _normalize_anthropic_url(url)
-    if provider == "ollama":
-        return _normalize_ollama_url(url)
-    if provider == "chatgpt-subscription":
-        return _normalize_chatgpt_subscription_url(url)
-    return _normalize_openai_chat_url(url)
-
-
 async def stream_llm(url: str, model: str, messages: List[Dict], temperature: float = LLMConfig.DEFAULT_TEMPERATURE,
                      max_tokens: int = LLMConfig.DEFAULT_MAX_TOKENS, headers: Optional[Dict] = None,
                      timeout: int = LLMConfig.STREAM_TIMEOUT, prompt_type: Optional[str] = None,
-                     tools: Optional[List[Dict]] = None, session_id: Optional[str] = None,
-                     tool_choice_none: bool = False, workload: str = "foreground"):
-    target_url = _stream_target_url(url)
-    async with _local_model_slot(target_url, model, workload):
-        async for chunk in _stream_llm_inner(
-            url,
-            model,
-            messages,
-            temperature=temperature,
-            max_tokens=max_tokens,
-            headers=headers,
-            timeout=timeout,
-            prompt_type=prompt_type,
-            tools=tools,
-            session_id=session_id,
-            tool_choice_none=tool_choice_none,
-        ):
-            yield chunk
-
-
-async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperature: float = LLMConfig.DEFAULT_TEMPERATURE,
-                            max_tokens: int = LLMConfig.DEFAULT_MAX_TOKENS, headers: Optional[Dict] = None,
-                            timeout: int = LLMConfig.STREAM_TIMEOUT, prompt_type: Optional[str] = None,
-                            tools: Optional[List[Dict]] = None, session_id: Optional[str] = None,
-                            tool_choice_none: bool = False):
+                     tools: Optional[List[Dict]] = None, session_id: Optional[str] = None):
     """Stream LLM responses with improved error handling.
 
     Yields SSE chunks:
@@ -2144,7 +1860,7 @@ async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperat
         h = _provider_headers(provider, headers)
         payload = _build_chatgpt_responses_payload(model, messages_copy, temperature, max_tokens, stream=True)
     else:
-        target_url = _normalize_openai_chat_url(url)
+        target_url = url
         payload = {
             "model": model,
             "messages": messages_copy,
@@ -2160,8 +1876,6 @@ async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperat
             payload[tok_key] = max_tokens
         if tools:
             payload["tools"] = tools
-        elif tool_choice_none:
-            payload["tool_choice"] = "none"
         # Mistral thinking-capable models — send reasoning_effort so Mistral
         # activates thinking mode and returns structured reasoning_content.
         # Effort level is configurable via ODYSSEUS_MISTRAL_REASONING_EFFORT
@@ -2174,7 +1888,6 @@ async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperat
         if _is_ollama_openai_compat_url(url) and _supports_thinking(model):
             payload["think"] = False
         _apply_local_cache_affinity(payload, url, session_id)
-        _apply_local_generation_stability(payload, target_url, model)
         h = _provider_headers(provider, headers)
         if provider == "copilot":
             from src.copilot import apply_request_headers
@@ -2191,7 +1904,6 @@ async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperat
         yield f'event: error\ndata: {json.dumps({"error": f"Upstream {_host_key(target_url)} unreachable (cooldown active)", "status": 503})}\n\n'
         return
     note_model_activity(target_url, model)
-    degenerate_guard = _DegenerateStreamGuard(model)
 
     # ── ChatGPT Subscription / Codex Responses streaming ──
     if provider == "chatgpt-subscription":
@@ -2226,10 +1938,6 @@ async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperat
                     if evt == "response.output_text.delta":
                         delta = data.get("delta") or ""
                         if delta:
-                            _degenerate = degenerate_guard.check(delta)
-                            if _degenerate:
-                                yield _degenerate
-                                return
                             yield f'data: {json.dumps({"delta": delta})}\n\n'
                     elif evt == "response.completed":
                         usage = (data.get("response") or {}).get("usage") or data.get("usage") or {}
@@ -2562,19 +2270,8 @@ async def _stream_llm_inner(url: str, model: str, messages: List[Dict], temperat
                                                 reasoning = (reasoning + thinking_part) if reasoning else thinking_part
                                             content = text_part
                                         if reasoning:
-                                            _degenerate = degenerate_guard.check(reasoning)
-                                            if _degenerate:
-                                                yield _degenerate
-                                                return
                                             yield _stream_delta_event(reasoning, thinking=True)
                                         if content:
-                                            content = _strip_visible_chat_template_artifacts(content)
-                                            if not content:
-                                                continue
-                                            _degenerate = degenerate_guard.check(content)
-                                            if _degenerate:
-                                                yield _degenerate
-                                                return
                                             content = re.sub(r"<mm:think(\s+[^>]*)?>", r"<think\1>", content, flags=re.IGNORECASE)
                                             content = re.sub(r"</mm:think>", "</think>", content, flags=re.IGNORECASE)
                                             stripped = content.lstrip()

src/model_context.py

@@ -220,10 +220,6 @@ KNOWN_CONTEXT_WINDOWS = {
     'hermes': 131072,
     'nous-hermes': 131072,
 
-    # --- Xiaomi ---
-    'mimo-v2.5-pro': 1048576,
-    'mimo-v2.5': 1048576,
-
     # --- Open community ---
     'dolphin': 32768,
     'mythomax': 4096,
@@ -316,83 +312,6 @@ def _lookup_known(model: str) -> Optional[int]:
     return best_ctx
 
 
-def _model_ctx_from_entry(m: dict) -> Optional[int]:
-    """Extract a positive context window from one /models catalog entry.
-
-    Checks the common top-level fields first, then a nested meta/model_extra
-    object. Returns None when no positive window is reported.
-    """
-    if not isinstance(m, dict):
-        return None
-    for field in (
-        "context_length",
-        "context_window",
-        "max_model_len",
-        "max_context_length",
-        "max_seq_len",
-    ):
-        val = m.get(field)
-        if val and isinstance(val, (int, float)) and val > 0:
-            return int(val)
-    meta = m.get("meta") or m.get("model_extra") or {}
-    if isinstance(meta, dict):
-        # n_ctx is the actual serving context (set via -c flag in llama.cpp)
-        for field in ("n_ctx", "context_length", "context_window", "max_model_len"):
-            val = meta.get(field)
-            if val and isinstance(val, (int, float)) and val > 0:
-                return int(val)
-    return None
-
-
-# Per-endpoint cache of the {model_id: context_length} map parsed from a
-# proxy/api catalog. api/proxy endpoints skip the /models download on every
-# lookup because a large catalog is expensive; caching the whole map lets us
-# pay that download at most once per endpoint instead of once per model.
-_catalog_ctx_cache: Dict[str, Dict[str, int]] = {}
-
-
-def _proxy_catalog_context(endpoint_url: str, model: str) -> Optional[int]:
-    """Context window for a model read from the endpoint's /models catalog.
-
-    Fetches the catalog once per endpoint and caches the full id->context map,
-    so an api/proxy endpoint serving a model that isn't in KNOWN_CONTEXT_WINDOWS
-    (e.g. a new OpenRouter model) still reports its real window instead of the
-    bare default. Returns None when the catalog can't be read or doesn't list a
-    positive window for the model.
-    """
-    cat = _catalog_ctx_cache.get(endpoint_url)
-    if cat is None:
-        from src.endpoint_resolver import build_models_url
-        try:
-            r = httpx.get(build_models_url(endpoint_url), timeout=REQUEST_TIMEOUT)
-        except Exception as e:
-            logger.debug(f"Failed to fetch proxy catalog for context length: {e}")
-            return None
-        if not r.is_success:
-            return None
-        cat = {}
-        try:
-            for m in (r.json().get("data") or []):
-                mid = m.get("id") if isinstance(m, dict) else None
-                ctx = _model_ctx_from_entry(m) if mid else None
-                if mid and ctx:
-                    cat[mid] = ctx
-        except Exception as e:
-            logger.debug(f"Failed to parse proxy catalog for context length: {e}")
-            return None
-        _catalog_ctx_cache[endpoint_url] = cat
-
-    if model in cat:
-        return cat[model]
-    # Catalog ids may carry a provider prefix (e.g. "openai/gpt-4o") while the
-    # session stores the bare id; match on the trailing segment as a fallback.
-    base = model.split("/")[-1]
-    for mid, ctx in cat.items():
-        if mid.split("/")[-1] == base:
-            return ctx
-    return None
-
-
 def _query_context_length(endpoint_url: str, model: str) -> Tuple[int, bool]:
     """Query the model API for context length. Returns (context_length, known) where
     ``known`` is False only for the bare DEFAULT_CONTEXT fallback."""
@@ -407,14 +326,6 @@ def _query_context_length(endpoint_url: str, model: str) -> Tuple[int, bool]:
         if known:
             logger.info(f"Using known context window for {model}: {known}")
             return known, True
-        # Not in the known table: read the real window from the catalog (cached
-        # once per endpoint) instead of capping every unknown model at the
-        # default — that under-reported large windows on aggregators like
-        # OpenRouter (issue #4886).
-        api_ctx = _proxy_catalog_context(endpoint_url, model)
-        if api_ctx:
-            logger.info(f"Proxy catalog reports context window for {model}: {api_ctx}")
-            return api_ctx, True
         return DEFAULT_CONTEXT, False
 
     # Try llama.cpp /slots endpoint first — reports actual serving context
@@ -455,7 +366,27 @@ def _query_context_length(endpoint_url: str, model: str) -> Tuple[int, bool]:
             for m in models_list:
                 mid = m.get("id", "")
                 if mid == model or mid.split("/")[-1] == model.split("/")[-1]:
-                    api_ctx = _model_ctx_from_entry(m)
+                    for field in (
+                        "context_length",
+                        "context_window",
+                        "max_model_len",
+                        "max_context_length",
+                        "max_seq_len",
+                    ):
+                        val = m.get(field)
+                        if val and isinstance(val, (int, float)) and val > 0:
+                            api_ctx = int(val)
+                            break
+
+                    if not api_ctx:
+                        meta = m.get("meta") or m.get("model_extra") or {}
+                        if isinstance(meta, dict):
+                            # n_ctx is the actual serving context (set via -c flag in llama.cpp)
+                            for field in ("n_ctx", "context_length", "context_window", "max_model_len"):
+                                val = meta.get(field)
+                                if val and isinstance(val, (int, float)) and val > 0:
+                                    api_ctx = int(val)
+                                    break
                     break
     except Exception as e:
         logger.debug(f"Failed to query context length for {model}: {e}")

src/model_discovery.py

@@ -187,10 +187,8 @@ class ModelDiscovery:
             r = httpx.get(f"{base}/models", timeout=3)
             if not r.is_success:
                 return None
-            data = r.json()
-            # Some OpenAI-compatible servers return a bare list, not {"data": [...]}.
-            items = data if isinstance(data, list) else ((data or {}).get("data") or [])
-            ids = [m.get("id") for m in items if isinstance(m, dict) and m.get("id")]
+            data = r.json() or {}
+            ids = [m.get("id") for m in (data.get("data") or []) if m.get("id")]
             if ids:
                 return {
                     "host": host,

src/personal_docs.py

@@ -68,8 +68,6 @@ def read_text_file(path: str) -> str:
 
 def split_chunks(text: str, size: int = config.CHUNK_SIZE, overlap: int = config.CHUNK_OVERLAP) -> List[str]:
     """Split text into overlapping chunks."""
-    if not isinstance(text, str):
-        return []
     text = text.strip()
     if not text:
         return []
@@ -89,8 +87,7 @@ def split_chunks(text: str, size: int = config.CHUNK_SIZE, overlap: int = config
 
 def tokenize(s: str) -> Set[str]:
     """Tokenize string into words, excluding stop words."""
-    text = s if isinstance(s, str) else ""
-    tokens = re.findall(r"[A-Za-z0-9_\-]+", text.lower())
+    tokens = re.findall(r"[A-Za-z0-9_\-]+", (s or "").lower())
     return set(t for t in tokens if t not in config.STOP_WORDS and len(t) > 1)
 
 def load_personal_index(