Mnemos is a powerful, persistent memory system designed for AI agents, ensuring they retain crucial information across interactions without relying on cloud services or GPUs. Operating entirely on CPU, it runs efficiently on standard hardware like laptops or Raspberry Pis. It offers a sophisticated hybrid retrieval pipeline, combining BM25, vector search, and cross-encoder reranking to accurately recall memories, all without an LLM in the search path. Integrates seamlessly with MCP-compatible AI clients such as Claude Code, Cursor, and ChatGPT Desktop, providing four essential tools for memory management, and includes a full CLI for independent use, keeping all your agent's memory secure and local.
Características Principales
01CPU-only operation for broad compatibility across various hardware.
02Native integration with MCP-compatible AI clients for seamless use.
03Full command-line interface (CLI) for flexible memory management.
04100% local data storage with no cloud dependencies or API calls.
05Hybrid retrieval pipeline (BM25 + vectors + cross-encoder rerank) for high accuracy.
061 GitHub stars
Casos de Uso
01Enhancing AI agent memory in MCP-compatible clients like Claude Code and Cursor.
02Managing and querying agent memories through a standalone command-line interface.
03Providing a private, local, and persistent knowledge base for AI systems.
