Manages and compacts an agent's local memory logs using an embedded database and AI embeddings.
Sponsored
Memory provides a local-first MCP server for agent memory, designed to separate fast capture, runtime retrieval, and maintenance curation processes. It uses SQLite with `sqlite-vec` for efficient embedded storage and search, leveraging OpenAI embeddings for semantic understanding. Agents can log candidate memory during their normal operations, which is then processed by a maintenance routine. This routine converts relevant log items into persistent, searchable facts, ensuring a structured and curated memory for AI agents.
Key Features
01Configurable OpenAI embeddings for semantic memory
02Separation of fast capture, runtime retrieval, and maintenance curation
030 GitHub stars
04Queue-based maintenance flow for memory logging and acknowledgement
05Embedded storage and search with SQLite + sqlite-vec
06CRUD tools for comprehensive memory management
Use Cases
01Logging candidate memory from AI agents during their operational tasks
02Enabling runtime retrieval of relevant facts for agent decision-making
03Curating and compacting agent memory through a structured maintenance process
