The model was hallucinating project names from task/note content (e.g.
inferring "Vehicle Maintenance" from "purchase wheel bearings"). Added
explicit guidance to both project field descriptions: only set if the
user explicitly named a project, never infer from content.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- build_context: when conversation_type is 'briefing', inject a system
prompt instruction telling the model to answer from conversation history
and article context instead of searching the web
- Consolidate briefing conversation type detection to one DB query (was
being checked twice — once for the system prompt addition, once for
article context injection)
- ChatPanel: render a visual 'New Briefing Update' separator line before
2nd+ briefing slot messages (identified by metadata.rss_item_ids)
- types/chat.ts: add metadata field to Message interface
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replaces the single monolithic research note with topic-driven section notes
plus an index note. Two new LLM calls: _generate_outline (JSON outline, 3-8
sections) and _synthesize_section (300-600 word focused note per section,
parallelised via asyncio.gather). Public signature of run_research_pipeline
unchanged; falls back to single-note synthesis on outline failure or if all
sections fail.
Also extracts _build_sources_block helper and adds full test suite.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Route now logs every synthesis request (char count, voice, speed)
- Route logs char count + text preview when the 8000-char limit is hit
- Route logs empty audio with preview (helps spot no-chunk-produced edge case)
- Route logs success with byte count and duration
- Kokoro synthesise() logs per-call: samples produced, elapsed, chars/s
- Kokoro synthesise() logs warning when zero audio chunks returned with preview
- Kokoro synthesise() catches and logs pipeline-internal errors with preview
- Frontend: console.warn now includes char count + 80-char preview on failure and retry
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
CONTENT_MAX_CHARS was removed from rss.py when the article content cap
was lifted. backfill_rss_article_content still referenced it, causing an
ImportError on startup.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When user_timezone is saved via PUT /api/settings, immediately call
update_user_schedule if briefing is enabled so the scheduler picks up
the new timezone without requiring a restart.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- _add_user_jobs now accepts config dict and skips disabled slots
- _get_briefing_enabled_users returns 3-tuple (user_id, tz, config)
- update_user_schedule passes config through to _add_user_jobs
- _run_slot_for_user skips morning slot on non-work days via profile.work_schedule
- Add tests for slot gating and work-day gate
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Removes the custom classify_capture_intent + _process_note two-pass
approach. The LLM now picks the right tool directly via Ollama's native
tool_calls API (same path as the main chat pipeline). _should_think
decides whether extended reasoning is needed based on input length/
complexity. intent.py deleted — no longer needed.
Android app and response format unchanged.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
POST /api/briefing/articles/<id>/discuss injects stored article content
as a persisted read_article tool exchange before triggering generation.
The LLM sees the article as already read; follow-ups retain context via
the fixed history builder. Frontend Discuss button now calls the new
endpoint instead of inlining article text in the user message.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Needed by the Discuss endpoint to persist synthetic read_article
tool exchanges before triggering generation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The history builder was silently dropping tool_calls from prior turns,
causing the LLM to lose article/search context on every follow-up.
Now reconstructs the assistant tool_call dict + per-call tool result
entries. Messages without tool_calls are unaffected.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Trafilatura extracts only article body text (typically 2K–15K chars),
so storing the full content is safe without an artificial ceiling.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Startup now pulls Config.OLLAMA_MODEL (system default chat model) — previously only
embedding and background models were pulled; the primary chat model was skipped
- _warm_user_models expanded to also pull user-configured default_model and
background_model overrides that are missing from Ollama, rather than logging and
skipping them; pulls run before warm/KV-cache priming
- Add background_model to _MODEL_KEYS in settings route so clearing the dropdown
deletes the DB row instead of saving "", which caused Ollama failures in tag
suggestions, title generation, project summaries, and RSS classification
- Add http/https scheme validation to PUT /api/admin/base-url matching the CalDAV
route pattern; a bad value no longer silently breaks invite/password-reset links
- Update admin voice config description: "Reload models" button exists to avoid
a server restart, so the old "restart required" text was misleading
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- PUT/PATCH/DELETE /api/tasks/:id now use get_note_for_user + can_write_note
so shared project editors can mutate tasks; owners unaffected
- PATCH /api/tasks/:id/status gets same treatment
- All write routes call update_note/delete_note with note.user_id (owner)
not the accessing user's uid, matching the milestone fix pattern
- create_task tool gains tags (array) and status (enum) parameters;
handler now passes tags to create_note and respects initial status
- create_task tool response now includes milestone_id and parent_id
- update_note tool gains milestone parameter; handler resolves the
milestone by title within the note's current (or newly set) project,
clears milestone_id when project is cleared
- list_tasks tool gains q keyword search parameter; passed through
to list_notes
- TaskEditorView: replace window.location.reload() with
router.push('/tasks/:id') after save
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Project.to_dict() now includes user_id and auto_summary
- Status validation unified to (active/completed/archived) on both
create and update project routes; update route previously had none
- Milestone routes: replace get_project (ownership-only) with
get_project_for_user so shared viewers/editors can access milestones
- Add get_milestone_in_project() to milestones service for project-
scoped lookup without user_id filter; all milestone routes use it
- Milestone PATCH now validates status as 'active'|'done'; fix tool
enum which was wrongly ['active','completed','cancelled']
- Write mutation routes (POST/PATCH/DELETE milestones) now check
can_write_project() and return 403 for read-only shared users
- update_project tool now exposes title and color fields so projects
can be renamed or recolored via chat
- create_project tool now exposes color field
- GET /api/projects?include_summary=true embeds summaries in one
backend pass; ProjectListView switches to this, eliminating N+1
per-project fetches
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- cleanup_old_conversations now excludes briefing conversations (was
silently deleting briefing history after the retention window)
- list_conversations response now includes rag_project_id, matching the
shape returned by the single-conversation GET endpoint
- create_conversation_from_article: removed duplicate async_session import
(_session2 was a copy of the same import); consolidated into one
- MAX_TOOL_ROUNDS fixed from 5→6 to match the actual range(6) loop;
loop updated to range(MAX_TOOL_ROUNDS) so the constant is accurate
- Chat retention cleanup moved from per-request (every GET /conversations)
to a daily scheduled job in event_scheduler.py; route no longer runs
a DB write on every read
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- RRULE expansion: list_events now expands recurring events into
individual occurrences within the query window using python-dateutil
- CalDAV pull sync: new caldav_sync.py + POST /api/events/sync route;
imports remote events into the internal store by caldav_uid
- Past event search: search_events accepts include_past=true to search
historical events; exposed in the LLM tool definition
- Internal reminders: migration 0037 adds reminder_minutes +
reminder_sent_at columns; event_scheduler.py checks every 5 min and
fires push notifications; CalDAV sync job runs hourly
- reminder_minutes now stored and returned in create/update routes + tools
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend:
- tools.py: apply UTC normalization to update_event datetime fields
(matched create_event which already did this)
- events.py service: allow end_dt/recurrence/project_id to be cleared
via update_event by permitting None for nullable fields
- events.py service: find_events_by_query now returns upcoming events
first, falling back to past — prevents AI tools from mutating stale
past events when a future match exists
- events.py service: list_events now uses overlap logic (start <= to
AND end >= from) so multi-day events spanning the query boundary
are included; previously only start_dt was checked
Frontend:
- ToolCallCard: fire fable:calendar-changed on created/updated/deleted
so CalendarView refetches without requiring a manual page refresh
- KnowledgeView: replace raw apiGet('/api/events') with listEvents()
client function; also fix today bar which was reading .events off a
flat array (always empty) — now correctly receives EventEntry[]
- HomeView: use full ISO strings for event date range instead of naive
UTC-midnight strings; deduplicate inline date math via _dateRange()
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Backend:
- GET /api/knowledge/ids: returns up to 100 note IDs cheaply (no body
parsing), supports same filters as /api/knowledge, includes has_more
- GET /api/knowledge/batch?ids=...: fetches full items for given IDs in
order; used by frontend to load content in controlled batches
Frontend (KnowledgeView):
- Fetch 100 IDs upfront, load first 50 as content on mount
- IntersectionObserver sentinel (root: null) triggers 24-item content
batches as user scrolls
- Proactive ID refill when queue drops below 48 unloaded IDs
- fetchGen counter invalidates stale in-flight responses on filter reset
- IDs claimed before async fetch to prevent double-loading
- sentinelVisible ref drives post-load re-check when content doesn't
push sentinel off screen
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
All datetime parsing now uses _parse_dt() which adds UTC tzinfo when
none is present, matching the fix already applied in tools.py. This
prevents asyncpg errors when comparing naive datetimes against
TIMESTAMPTZ columns — the root cause of events not appearing in the
calendar view.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
KnowledgeView: sentinel was OUTSIDE the card-grid div, making
IntersectionObserver (root: cardGridEl) never fire since the target must
be a descendant of the root. Moved sentinel inside card-grid with
grid-column:1/-1 + order:9999 so it spans all columns and sits at the
bottom. Fixed backup check to compare against container bounds not viewport.
tools.py list_events: apply same UTC normalization as create_event (treat
naive datetimes as UTC, handle Z suffix). Update tool description to
explicitly request full-day UTC ranges so the LLM doesn't send local time
without offsets, which caused the recall query to miss UTC-stored events.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- KnowledgeView: IntersectionObserver was watching the viewport instead
of the card-grid scroll container, causing infinite scroll to stop
loading after only ~29 items. Pass card-grid element as `root`.
- CalendarView: listen for 'fable:calendar-changed' custom event and
call refetchEvents() so tool-created events appear without navigation.
- ChatPanel: dispatch 'fable:calendar-changed' when create_event,
update_event, or delete_event tool calls succeed.
- tools.py: normalize naive datetimes to UTC before storing events so
timezone comparisons in list_events queries are always consistent.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds pick_num_ctx() which selects the smallest context window tier
(8192, 16384, 32768) that fits the current messages with 25% headroom,
capped at OLLAMA_NUM_CTX. Threads num_ctx through generation_task.py so
every chat request uses the computed tier rather than a fixed 16384.
Fixes a critical cache miss bug: KV cache priming in app.py and
settings.py was sending requests without num_ctx, so Ollama sized the
cache at its model default (different from the 16384 real requests used),
forcing a full model reload on the first real user message. Both priming
sites now call pick_num_ctx() and pass the matching value.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Removes verbose redundant text from tool descriptions and system prompt
guidance: multi-line recurrence_rule JSON examples, CAPS warnings that
duplicate system prompt instructions, and wordy descriptions that don't
add model understanding.
Saves ~990 tokens per request (~17% reduction, 5,639 → ~4,650 tokens),
reducing prefill time on cache misses and lowering KV memory pressure.
No functional changes — parameter names, types, enums, and required
fields are unchanged.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When a user saves a new default_model in Settings, fire a background
cache-prime request so the first message with the new model is fast
rather than paying the full cold-start prefill cost.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
After loading each user's chat model into VRAM, send a minimal chat
request with the real system prompt (num_predict=1) to populate the
KV cache. The first real user message then only needs to process its
own tokens rather than the full ~5,600-token system prompt, reducing
cold-start TTFT from ~25s to <1s.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Exposes OLLAMA_BACKGROUND_MODEL as a per-user setting in General settings,
alongside the Chat Model selector. Includes an inline warning when the same
model is selected for both, explaining the KV cache performance impact.
All background task callers (title generation, tag suggestions, project
summaries, RSS classification) now read background_model from user settings,
falling back to OLLAMA_BACKGROUND_MODEL env var.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Background tasks (title generation, tag suggestions, project summaries,
RSS classification) were using qwen3:8b and wiping its KV cache after
every response, preventing prefix cache hits on subsequent user messages.
Adds OLLAMA_BACKGROUND_MODEL (default: qwen2.5:0.5b) config var and
routes all background LLM calls to it, keeping qwen3:8b's KV cache
warm between user messages for consistent sub-second TTFT.
Also adds infinite scroll to KnowledgeView (replaces load-more button)
and bakes spaCy en_core_web_sm into the Docker image to eliminate the
pip install on every startup.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
get_weather now returns 1 day by default (today) instead of a full 7-day
forecast. A new optional `days` parameter (1–8) lets the model request
more days when the user explicitly asks for a weekly forecast or specific
date range. Tool description updated to guide the model accordingly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- KnowledgeView minichat: render assistant messages through renderMarkdown
so headers, bold, lists etc. display correctly instead of raw markdown
- get_weather tool: read user's temp_unit from briefing_config and convert
temperatures to °F when preferred; also include temp_unit in the
returned payload so the model can label values correctly
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- KnowledgeView minichat: add margin-inline: auto so the widget centers
within the content area when max-width is reached on wide screens
- weather get_weather tool: return success: true on both the arbitrary
location path and the cached locations path so ToolCallCard shows
the correct success state instead of always flagging as error
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
RAG notes, RSS news, current note, URL content, and briefing articles
are now prepended to the user message rather than appended to the system
message. The system message now contains only stable content (persona,
tool guidance, date, profile, workspace, history summary), making its
token sequence identical across consecutive requests and allowing
Ollama's KV prefix cache to fire reliably every time.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Routes simple/conversational messages to think=false automatically,
even when the user has thinking enabled. Patterns checked: word count
thresholds, complexity keywords, code blocks, skip patterns for greetings
and simple CRUD. Workspace mode (think=true from frontend) still benefits
from the classifier on short messages.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
wait_for_model_loaded() polled /api/ps for up to 180s waiting for the
model to appear as loaded. But Ollama lazy-loads models on the first
/api/chat request, so the poll will never succeed — it just blocks for
the full 180s after every Ollama restart before proceeding.
Removed the wait entirely. Ollama handles on-demand loading correctly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
65536 was causing Ollama to allocate a ~50GB KV cache, spilling 77% of
the model to CPU RAM and making prefill extremely slow (35-125s TTFT).
16384 covers 30+ message conversations comfortably while keeping the KV
cache small enough to stay on GPU.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move static content (persona + tool guidance) to a fixed prefix and
append all dynamic content (date, timezone, profile, entities) as a tail.
Ollama prefix caching requires byte-for-byte token match from the start
of the prompt. Previously, Today's date + user profile were embedded
mid-prompt, invalidating the cache on every request/day and causing
~20s TTFT regardless of model warmth.
With this change the static prefix (~5500 tokens) should be cached
after the first request each session, reducing TTFT to ~2-5s for the
~200-token dynamic tail only.
Also removed inline user_timezone from tool_lines (timezone is now
stated once in the dynamic tail, which the model reads).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
bvandeusen