Search memories by semantic + keyword query, or bulk-fetch by ID.

When ids is provided, skips the vector/BM25 pipeline entirely and does a direct SQL lookup by lore_id. query is ignored in that path.

Insert memories and/or links into the store.

Each memory dict must include:

• title (str, required): short unique label for the memory

• content (str, optional): the full text to store

• description (str, optional): brief summary

• score (float, optional, default 5.0): initial quality score 0-10

• source_type (str, optional, default 'observed'): provenance tag. One of: observed, inferred, user_stated, consolidated, injected.

• links (list[dict[str, Any]], optional): inline links to create after insert. Each link dict: {target_memory_id (str, required), relation_type (str, required), reason? (str)}

Each top-level link dict must include source_memory_id, target_memory_id, relation_type, and reason.

Suggest link candidates between a memory and related memories. Returns ranked candidates with per-signal scores. Does NOT write any links — call lore_insert with links=[] to confirm.

Capture a thought instantly — one fact, one call.

Use this when you discover something worth keeping: a decision, a bug root cause, a user preference, a pattern.

Minimal effort, high reward. Your future self will find this useful.

Rate memories and links after using them. Drives the quality signal loop.

Each memory_feedback dict: {id (str), useful (bool), confidence (int 1-10)}. Each link_feedback dict: {id (str), useful (bool), confidence (int 1-10)}.

useful=True bumps score; useful=False deducts. Confidence scales the delta. Repeated useful=False with low confidence triggers soft-delete. Call after every lore_search to keep scores calibrated.

Return all session IDs that have been marked as processed via lore_reflect.

Reflect on a completed session — save what you learned.

Minimal usage: pass session_id and summary. That's enough. The rest are extras for when you discovered something substantial.

Soft-delete one or more memories by ID.

Memories are marked soft_deleted=1 and excluded from future search results. This is reversible at the DB level but no undelete tool is exposed in v1.

Retrieve pending link suggestions for review, sorted by quality score.

Returns the top candidates from the sweep engine's pending queue. Use lore_review_suggestion to accept or reject them.

Accept or reject one or more link suggestions in a single call.

Processes each suggestion independently — a failure on one does not block the rest. Suggestion rows are never deleted; status is updated to 'accepted' or 'rejected' for audit trail.

On accept: creates a real memory_links row using the suggestion's suggested_type (falls back to 'references' if None or unrecognised).

On reject: marks the suggestion as rejected. Future sweeps skip this pair.

Idempotent per item: double-accept and double-reject both return status='skipped' with an explanatory message.