Create new entities in the knowledge graph.

Each entity dict has: name (str), entityType (str), observations (list[str]). Optional: project (str). Duplicates are silently ignored.

Add new observations to existing entities.

Each dict has: entityName (str), contents (list[str]). Duplicate observations are silently ignored.

Create relations between entities in the knowledge graph.

Each dict has: from (str), to (str), relationType (str). Duplicate relations are silently ignored.

Delete entities and their associated observations and relations (CASCADE).

Also cleans up the FTS index.

Delete specific observations from entities.

Each dict has: entityName (str), observations (list[str]).

Delete specific relations from the knowledge graph.

Each dict has: from (str), to (str), relationType (str).

Read the full knowledge graph with pagination.

Returns JSON: {entities: [{name, entityType, observations: [...]}], relations: [{from, to, relationType}], total: int, has_more: bool}

Search the knowledge graph using hybrid BM25 + semantic search.

When sqlite-vec is installed, combines FTS5 keyword matching with vector cosine similarity via Reciprocal Rank Fusion. Falls back to FTS5-only otherwise. Results are re-ranked with 6 contextual signals (recency, project affinity, graph proximity, richness, canonical facts, session).

Open specific entities and retrieve their inter-relations.

Returns the requested entities with observations and all relations that exist between them.