mnemosyne-memory-provider

Set up a local memory backend by using fastembed for embedding generation and sqlite-vec for vector search. This provides efficient memory retrieval directly on Linux and macOS platforms without external API dependencies.

Memory consolidation automates the conversion of working memories into episodic summaries. You need it when your agent accumulates excessive interaction data, requiring summarization to maintain recall performance and reduce storage overhead.

Fastembed works with sqlite-vec to provide local vector search on macOS. This stack utilizes Python libraries and numpy for data processing, ensuring embedding generation and retrieval without requiring dedicated external database servers.

Troubleshoot missing embeddings and database locks by following the built-in diagnostic guide. It details steps to resolve recall problems, database lock contentions, and embedding generation failures within the local memory backend.

Limitations of using sqlite-vec include potential database locks during concurrent access and platform optimization restricted to Linux and macOS. It is not designed for distributed environments, relying entirely on local Python resources.

You need numpy along with fastembed and sqlite to run the memory provider. These dependencies support local embedding generation, vector search optimization, and the underlying numerical data processing.