Provides AI agents with a comprehensive, local-first cognitive memory engine featuring namespace-isolated vector storage, advanced retrieval mechanisms, and knowledge graph capabilities.
Engram Memory serves as a powerful cognitive memory engine specifically designed for AI agents, operating as a local-first MCP server. It offers namespace-isolated vector storage, enabling LLMs to maintain distinct memory contexts. Its sophisticated retrieval system combines vector search with BM25 keyword matching, enhanced by features like synonym expansion and physics-based re-ranking to deliver high recall. Beyond simple storage, it includes a knowledge graph for semantic linking, intelligent clustering for organization, and lifecycle management with activation energy decay, allowing memories to evolve and fade based on relevance and usage. The engine also supports multi-agent sharing and hierarchical expert routing, making it a versatile and intelligent memory solution for complex AI systems.
Key Features
01Semantic clustering (DBSCAN) and multi-agent namespace sharing
02Namespace-isolated vector storage with Int8 scalar quantization
03Lifecycle management with activation energy decay and physics-based re-ranking
