graph-recall

Long-term memory retrieval for AI agents is implemented by combining vector-based semantic search with recursive graph traversal on a SQLite database to identify and expand context for entities, incidents, and historical threads.

Context expansion automatically enriches user queries by retrieving relevant 1-hop and 2-hop graph connections, pulling in related system components and prior decisions to provide a comprehensive, grounded response.

Yes, semantic graph recall requires a pre-populated memory database and the sqlite-vec extension to perform high-precision semantic lookups and recursive graph traversal effectively.

The best way to structure unbounded memory growth is using a graph-based retrieval system that maintains semantic relationships between facts in a SQLite-backed knowledge graph, preventing disconnected memory expansion.

Yes, you can use sqlite-vec for vector-based similarity searches combined with recursive graph traversal to find related entities and facts within a SQLite-backed knowledge graph.

Graph-based retrieval requires a pre-populated memory database to function, meaning it cannot perform semantic lookups or context expansion on empty datasets or without the sqlite-vec extension installed.