- ensure_collection was called without await, so the coroutine never ran
and fresh tenants had no collection when the upsert hit Qdrant
- a single None entry from embed_batch poisoned the point batch and
aborted the whole document upsert; failed chunks are now skipped with
a warning (all-failed raises and the IndexResult reports failure)
- raw client.delete/client.upsert calls now go through the async wrapper
(delete_by_filter and the new batch upsert_points method)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QbFZyDvYksazX6nYQYZ67L
Every vector call ran on the sync QdrantClient inside async wrapper
methods, blocking the FastAPI event loop per Qdrant round-trip. The
wrapper now holds an AsyncQdrantClient (timeout via QDRANT_TIMEOUT,
default 30s) and awaits all client calls; the wrapper API is unchanged.
Call sites off the wrapper were fixed too: the HybridRAG document leg
now uses the async search_vectors wrapper instead of the deprecated raw
client.search (also fixing its call to the nonexistent ollama.embed_text
which made the leg permanently report 'failed'), the health check awaits
get_collections, and document_sync's raw delete/upsert calls are awaited
(routed through wrappers in the next commit).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QbFZyDvYksazX6nYQYZ67L
ROOT CAUSE (investigated read-only against prod): the production
container sets OLLAMA_MODEL=nomic-embed-text (the embedding model), which
the pre-rename generation setting also read, so every consolidation
/api/generate call failed with HTTP 400 ('"nomic-embed-text" does not
support generate' - confirmed in prod logs and by a direct Ollama probe).
_classify_web_results_unified swallowed that as an empty classification,
and consolidate_knowledge marked EVERY SearchQuery processed anyway -
permanently draining the queue with zero pages ever created. Live Neo4j
shows 197/200 SearchQuery nodes processed=true with no output; every
subsequent 30-minute run then logged 'No unprocessed searches found'.
The label/tenant scoping was NOT at fault: persistence writes both the
tenant label and the plain :SearchQuery label the loop matches on.
The model resolution itself was already fixed in Phase A (94482bc,
ollama_llm_model / OLLAMA_LLM_MODEL). This commit repairs the pipeline
defect that masked it:
- LLM infrastructure failure (no output from generate) now raises
ConsolidationLLMUnavailableError instead of returning an empty routing
- consolidate_knowledge leaves those searches UNPROCESSED for the next
run, aborts the rest of the batch (the LLM is down for all of them),
and reports searches_deferred
- unparseable-but-present model output is still consumed (avoids
retrying a bad prompt forever); low-web skips unchanged
- every run logs 'Consolidation run complete: searches_processed=N
searches_deferred=M duration_ms=X'; both fields added to the response
- lookback boundary is now timezone-aware UTC (Neo4j datetime() reads
naive strings as UTC, shifting the window on CET hosts)
10 new offline regression tests.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QbFZyDvYksazX6nYQYZ67L
Replace the four stub endpoints with real implementations, all requiring
an explicit tenant user (Phase B rule):
- /ingest/check-updates: GraphService now records a SHA-256 content_hash
on every Document node at ingestion time; the endpoint compares those
stored hashes against current Wiki.js page content in one UNWIND Cypher
query per tenant and returns changed/new/deleted page lists (entity-stub
pages excluded, pre-hash-tracking documents flagged stored_hash_missing).
- /ingest/status/{job_id}: backed by the Redis JobManager; jobs are
tenant-scoped (foreign jobs 404). /ingest/page, /ingest/batch and
/ingest/all now create job records and return job_id.
- /ingest/repo-status/{repository}: wiki page count vs indexed Document
nodes under users/{tenant}/{repository} plus tenant job stats.
- /deduplicate/check: tenant-scoped Qdrant similarity scan; chunk pairs
above ~0.9 cosine from different pages grouped per page pair with best
score and page references (read-only).
Supporting changes: get_job_manager dependency (+ shutdown close),
scroll_all_points can return vectors, VectorService.find_duplicate_pairs,
src/core/hashing.compute_content_hash. 13 new offline unit tests.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QbFZyDvYksazX6nYQYZ67L
A live /query/hybrid probe as user=llm_tester returned jpmschweitzer
pages. Audit of all legs (vector, graph, web-persistence, volatile,
documents) plus enrichment/persistence found and fixed these unscoped
paths:
- vector_service.update_from_page and graph_service.update_from_page now
refuse pages outside users/{user}/ - previously any tenant could
ingest any wiki page (incl. another tenant's) into its own collection
and graph labels, which is how foreign content entered the vector leg.
- ingestion_service.ingest_all_pages clamps path_prefix to the caller's
namespace (segment-exact, sanitized comparison) and defaults to
users/{user}; /ingest/all returns 400 on cross-tenant prefixes.
- hybrid_rag_service._persist_search_for_librarian linked SearchQuery
nodes to unscoped (d:Document {page_id}); now matches only
User_{Tenant}_Document nodes.
- graph_service: _get_entity_mention_count, entity-stub mention/related
queries, generate_entity_stubs, find/purge_orphan_entities matched
unscoped Document nodes; cleanup_broken_relationships matched all
tenants' SearchQuery nodes; _entity_has_wiki_page listed all wiki
pages. All are now tenant-label / namespace scoped.
- volatile_service collection names now use the sanitized user id.
- is_path_in_user_namespace enforces a path-segment boundary
(users/llm_tester2 is not llm_tester's namespace) and treats
hyphen/underscore tenant spellings as the same sanitized tenant.
- New offline unit tests per leg (mocked clients) assert the
tenant-scoped collection/label/path is used and cross-tenant access
is refused.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
/query/graph scoping was a documented no-op (graph_service returned the
query unscoped) and neo4j_client permitted writes; a live probe showed a
nonexistent user could read the whole graph.
- Add Neo4jClient.execute_read() that opens the session with
default_access_mode=READ_ACCESS so the database refuses writes even if
validation is bypassed.
- GraphService.execute_query() now rejects queries containing
CREATE/MERGE/DELETE/DETACH/SET/REMOVE/DROP/FOREACH/LOAD or any CALL
(conservative word-boundary denylist on the uppercased query) and
executes through the read-only session; the no-op _scope_query_to_user
is removed.
- Remove the false user-scoping claims from /query/graph (main.py) and
/graph/query docs and the CypherQueryRequest model: the endpoints are
documented as admin/debug, unscoped read-only (per-tenant label
injection for arbitrary Cypher would need a real parser; /graph/nodes
remains the tenant-scoped path).
- Offline unit tests: denylist coverage (incl. lowercase/multiline/CALL),
word-boundary false-positive check, and READ_ACCESS session assertion.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Remove the implicit jpmschweitzer default tenant (DEFAULT_USER) from
src/core/multi_tenancy.py and every endpoint and request model that
inherited it (~40 endpoints across /query, /wiki, /vector, /graph,
/ingest, /volatile, /documents, /stats, /rag).
- Add validate_required_user() + RequiredUser pydantic type in
multi_tenancy and a shared require_user FastAPI dependency
(RequiredUserQuery) that rejects missing, empty, and whitespace-only
users with 422, following the /maintenance/* pattern.
- Wiki page create / smart-create / dossier request models now require
user (no fallback in wiki_service).
- /maintenance/cleanup/test-data derives the tenant from the page path
instead of using the production tenant collection.
- Wiki.js change listener skips changes when no tenant user can be
derived from the notification email instead of defaulting to the
production tenant.
- Consolidation service internal helpers no longer default to the
production tenant.
- Tool catalog marks user as required with honest descriptions.
- OpenAPI descriptions updated honestly; CHANGELOG notes that callers
(tatlock, Scheduler ingest tasks) must now send explicit user.
- Offline tests: 422 coverage for query/body endpoints, required-user
validator tests; updated legacy tests that assumed a default tenant.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Each retrieval leg (vector, graph, web, volatile, documents) now returns
(results, timing, error) instead of swallowing exceptions to an empty list.
The response reports per-leg status ('ok'/'failed'/'disabled') in
source_status and sets degraded=true when any enabled leg failed. Failed
legs still contribute no results (behavior unchanged) and are logged at
WARNING. Both fields are additive with defaults, so deploy order relative
to consumers does not matter.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The deployed container sets OLLAMA_MODEL=nomic-embed-text for embeddings,
which shadowed the generation-model setting and broke Phase 0 keyword
extraction and Phase 4 LLM re-ranking on every request. The setting is now
ollama_llm_model (env: OLLAMA_LLM_MODEL, default gemma4:e2b), startup logs
the resolved generation model, and Phase 0/Phase 4 LLM calls are wrapped in
a 12s asyncio.wait_for with graceful fallback so a hung call cannot gate
retrieval for the full 120s client timeout.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- POST /volatile/fetch/environment/{city} fetches both in parallel
- Single geocode lookup shared between API calls
- Uses asyncio.gather() for concurrent external requests
- Fix scheduler executor name (rest_api → rest_api_executor)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add FORECAST namespace for multi-day outlook (12hr TTL)
- WEATHER namespace now stores only current conditions (1hr TTL)
- Split fetch_weather into fetch_current_weather + fetch_forecast
- Add POST /volatile/fetch/forecast/{city} endpoint
- Different update frequencies for efficient caching
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Weather fetch now returns 7-day forecasts with UV index
- New /volatile/fetch/sun/{city} endpoint for sunrise/sunset
- New /volatile/fetch/air_quality/{city} endpoint for AQI and pollutants
- OpenMeteoProvider now implements AirQualityProvider interface
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The _is_automated_user method was never called - loop prevention is
handled by debouncing instead. User email filtering was intentionally
removed because the notification email is the page CREATOR, not editor.
- Remove unused _is_automated_user method
- Update test to verify notifications are processed regardless of user
- Remove obsolete test_automated_user_filtering test
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add SchedulerClient to communicate with external scheduler service for
registering volatile prefetch tasks discovered during HybridRAG searches.
- Add scheduler_client.py with full REST API for task CRUD operations
- Add scheduler_url config setting (default: http://scheduler:8090)
- Update consolidation service to use scheduler for prefetch registration
- Add scheduler health checks to startup/shutdown lifecycle
When HybridRAG classifies web content as prefetch-worthy, it now creates
scheduled tasks that periodically refresh the volatile cache via the
external scheduler service.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Phase C of memory system: Documents are now a retrieval source alongside
wiki, volatile, and web search.
Changes:
- Add enable_documents, document_limit, document_threshold to HybridRAGConfig
- Add paperless_id field to HybridRAGResult
- Add document_ms timing to TimingBreakdown
- Add document search to parallel retrieval (filters doc_type=document)
- Update RRF fusion to include documents as fourth source
- Add document metadata (correspondent, document_type, tags) to results
HybridRAG now searches 4 sources in parallel:
- Wiki (vector + graph merged)
- Volatile cache (priority boost)
- Paperless documents (new)
- Web search
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add MemoryRouteClassification and MemoryRoutingResult models
- Implement unified classifier (_classify_web_results_unified) that routes
web results to: wiki, volatile, file (Paperless), prefetch, or skip
- Add routing methods: _route_to_volatile, _route_to_files, _register_prefetch
- Update _process_search to use unified classifier instead of separate analysis
- Add get_volatile_cache_service factory to dependencies
- Wire volatile_service and settings_client into ConsolidationService
- Update ConsolidationResult/Response with new routing counters
Test fixes:
- Fix WikiJSClient fixtures to use api_token instead of username/password
- Fix entity linking test assertions to expect full user-namespaced paths
- Add sample_unified_classification fixture for new classifier format
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add VolatileFetchService to orchestrate API fetch and cache storage
- Add POST /volatile/fetch/weather/{city} endpoint
- Add POST /volatile/fetch/news/{category} endpoint
- Add POST /volatile/fetch/stock/{symbol} endpoint
- Add POST /volatile/fetch/crypto/{symbol} endpoint
Endpoints integrate with external API providers (OpenMeteo, NOS/BBC,
AlphaVantage) and store results in volatile cache with configurable TTL.
Designed for scheduler cron jobs to prefetch user-relevant data.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
get_all_chunk_references was missing paperless_id field needed for
Paperless orphan detection.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- POST /maintenance/cleanup/paperless - detect and clean orphaned Paperless documents
- Checks indexed documents against Paperless API
- Removes vectors and graph nodes for deleted documents
- Supports dry_run mode for preview
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Paperless API requires field ID (integer) not field name (string)
when updating custom fields. Now looks up field ID by name before
updating library_indexed custom field.
Also includes webhook debugging endpoint for development.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Change model field from document_id to id (Paperless sends id)
- Add content, created, modified, added, original_file_name, owner fields
- Add extra="ignore" config to handle additional Paperless fields
- Update sync service to use content from webhook payload
- Skip Paperless API call when content already provided
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add /documents router with webhook, upload, search, health endpoints
- Create DocumentSyncService for indexing documents to vectors/graph
- Add PaperlessClient for REST API integration
- Configure dependency injection for Paperless client
- Add document models for webhook payloads and responses
- Event-driven architecture via Paperless workflow webhooks
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Migrate volatile backend from Redis to Qdrant for semantic search
- Add natural language conversion for structured data embedding
- Simplify API: /volatile/search, /volatile/store, /{namespace}/{key}
- Integrate volatile into HybridRAG with priority boost in RRF fusion
- Add POST /maintenance/cleanup/volatile for expiry purging
- Update tests for new Qdrant-based architecture (37/37 pass)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Complete maintenance subsystem for index health and cleanup:
Endpoints:
- GET /maintenance/health - lightweight (or detailed) health check
- POST /maintenance/cleanup/all - full orphan cleanup
- POST /maintenance/cleanup/vectors - purge orphan vector chunks
- POST /maintenance/cleanup/graph - purge orphan graph nodes
- POST /maintenance/reconcile-index - cleanup + reindex missing pages
Bidirectional orphan detection:
- find_documents_without_vectors() in GraphService
- find_chunks_without_graph_nodes() in VectorService
Redis integration:
- Tracks last_cleanup timestamp for scheduler visibility
Config additions:
- Document store, volatile cache, and maintenance settings
- VectorServiceDep and GraphServiceDep type aliases
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add OLLAMA_EMBEDDING_MODEL for embeddings (nomic-embed-text)
- OLLAMA_MODEL now used for all LLM operations (mistral-nemo-large:latest)
- Remove separate reranker_model setting
- Update WikiPageWriter to use settings instead of hardcoded model
- Improves VRAM efficiency by keeping one model hot
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Merge vector+graph into single wiki source before RRF with web
- Wiki pages no longer get 2x advantage from dual retrieval
- Add vector similarity threshold (0.7 default)
- Skip synonyms in graph search to reduce noise
- Fix duplicate entity links bug in graph search
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add /rag/search endpoint for web, news, and image search via SearXNG
- Add /content/extract and /content/extract/batch endpoints
- Add ContentExtractor client using Trafilatura for content extraction
- Enhance HybridRAG web search with full content extraction
- Add Redis caching for search results
- Add new configuration options for search and extraction timeouts
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add POST /wiki/pages/smart-create endpoint that combines research with
content generation for the librarian agent:
- Run HybridRAG search on topic (wiki + graph + web)
- Use LLM (WikiPageWriter) to synthesize findings into wiki content
- Create page with proper attribution and sources
- Schedule background tasks for vector/graph indexing
- Apply bidirectional entity linking (forward + backward links)
New files:
- src/services/entity_linking_utils.py - shared entity linking helper
Modified:
- src/models/wiki.py - WikiSmartCreateRequest/Response models
- src/services/wiki_service.py - smart_create_page() method
- src/routers/wiki.py - /pages/smart-create endpoint
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
