Open-source memory infrastructure for AI agents.
open sourceA self-hostable memory layer. It ingests text, files and URLs, uses an LLM to extract atomic memories, indexes them for semantic search and hands agents the right context back through a REST API or an MCP server.
highlights
- 01Extraction engine with pronoun/entity resolution, memory versioning and contradiction handling.
- 02Vector search with reranking, multi-tenant isolation and an entity-relationship graph.
- 03OpenAPI spec, scoped API keys, rate limiting and encrypted settings.
- 04MCP server package, Docker and docker-compose, and a deployment guide.
- 05CI (type checks, dependency audit, build, Docker build), about 87 unit and integration tests and LoCoMo benchmark harnesses.
built with
- typescript
- hono
- convex
- embeddings
- mcp
- docker
- vitest