Interface MemoryConfig

Persistence backend for memory traces.

The Phase 1 facade currently implements the SQLite path at runtime. Other values are reserved for future backends and will throw if selected.

• 'sqlite' – file-based SQLite (implemented; recommended).
• 'memory' – reserved for a future in-process backend.
• 'qdrant' – reserved for a future vector-database backend.
• 'neo4j' – reserved for a future graph-database backend.
• 'hnsw' – reserved for a future ANN-only backend.

Default

'sqlite'
Copy

File-system path for stores that require one (e.g. SQLite db file). Ignored by in-memory and remote stores.

Example

'./data/agent-memory.sqlite'
Copy

Postgres connection string. Required when store='postgres'.

Example

'postgresql://postgres:wunderland@localhost:5432/agent_memory'
Copy

Qdrant base URL. Required when store='qdrant'.

Example

'http://localhost:6333'
Copy

Qdrant API key for cloud instances. Optional.

Embedding model configuration (provider name + optional model).

Optional embedding function for generating vectors at remember/recall time. When provided, enables HNSW vector search in recall() and stores embeddings alongside text in remember(). Without this, recall() falls back to FTS5-only.

Example

const mem = await Memory.create({
  embed: async (text) => {
    const res = await openai.embeddings.create({ model: 'text-embedding-3-small', input: text });
    return res.data[0].embedding;
  },
});
Copy

Whether to build and maintain a knowledge graph alongside the vector store. When enabled, entity co-occurrence and semantic edges are tracked.

Default

false
Copy

Whether the agent may autonomously refine and restructure its own memories (write new insight traces, prune contradictions, merge redundancies).

Default

false
Copy

Whether memory traces lose strength over time following an Ebbinghaus forgetting-curve model.

Default

true
Copy

Consolidation schedule and thresholds.

Document ingestion settings applied to all ingest() calls by default.