Provides a local, offline memory server for AI agents, leveraging SQLite with advanced search capabilities.
Memento is a robust, fully-offline MCP (Memory & Context Protocol) memory server designed to enhance the capabilities of AI agents. It utilizes SQLite for persistence, FTS5 for fast keyword searches, and `sqlite-vec` with an integrated `bge-m3` embedding model for powerful semantic vector search. This combination enables the creation of a structured knowledge graph of entities, observations, and relations, ensuring that AI models can maintain persistent, context-rich memories for improved performance and consistency in their interactions.
Key Features
01Offline `bge-m3` embedding model via @xenova/transformers
02Easy integration with Claude Desktop via MCP
030 GitHub stars
04Fast keyword search via FTS5
05Semantic vector search using `sqlite-vec` (1024d)
06Structured graph of entities, observations, and relations
Use Cases
01Integrating custom memory backends with MCP-compatible AI interfaces
02Enhancing LLM context and memory management
03Building persistent knowledge bases for AI assistants
