Dense-Mem

•
40
•

Provides a self-hosted memory layer for AI agents, featuring evidence provenance, conflict detection, and structured recall.

Dense-Mem is a self-hosted AI agent memory server designed to provide a durable and trustworthy memory layer. It ensures evidence provenance, manages typed claims, detects and handles conflicts without silently overwriting facts, and facilitates structured recall. This system empowers AI agents with reliable memory by enforcing verification gates and promotion rules, while allowing operators to maintain full control over data storage, team isolation, and API access. It functions as a standalone HTTP Model Context Protocol (MCP) memory server, returning structured outcomes that host LLMs can explain to users. Dense-Mem is integral to governing enterprise AI memory beyond traditional RAG approaches, leveraging knowledge graphs.

주요 기능

01Self-hosted deployment with team isolation, API key management, and audit metadata
021 GitHub stars
03Server-side embeddings and structured recall of facts, claims, fragments, and contradictions
04Evidence provenance with typed claims, verification, and promotion gates
05Comparable conflict detection returning clarification tasks, preventing silent fact overwrites
06Durable storage utilizing PostgreSQL for data and Neo4j for knowledge graphs

사용 사례

01Managing real-time memory insertions and updates for AI agent chat sessions
02Retrieving verified facts, claims, and clarifications to inform agent responses and decision-making
03Ingesting and structuring historical conversations into a persistent, verifiable knowledge base