Cozo Memory provides a robust, local-first memory system specifically designed for AI agents, built around an embedded CozoDB instance within Node.js. It distinguishes itself by offering a unique hybrid search that combines semantic vector embeddings, full-text search, and advanced graph algorithms (like PageRank and Shortest Path) fused via Reciprocal Rank Fusion (RRF). This tool ensures persistent, traceable memory with features such as time-travel queries, near-duplicate detection, multi-vector support for nuanced entity representation, and an intelligent Janitor service for knowledge consolidation. Operating entirely locally without cloud dependencies, Docker, or external services, Cozo Memory is ideal for developing intelligent, self-contained AI applications that require sophisticated context retrieval and knowledge management.
主な機能
01Graph-RAG & Graph-Walking for advanced context retrieval
02Time-Travel Queries with CozoDB Validity for historical data
030 GitHub stars
04Hybrid Search (Vector, Full-Text, Graph with RRF)
05Multi-Vector Support (Content-Embedding, Name-Embedding) for entities
06Semantic & Persistent Caching (L1/L2) with Janitor cleanup
