Multi-Agent Memory provides a centralized "brain" for AI agents, allowing them to store, recall, and share information seamlessly across different systems like Claude Code, OpenClaw, and n8n. It addresses the common problem of agents losing context between sessions or failing to learn from each other's discoveries. Unlike traditional solutions, it offers typed memory (events, facts, statuses, decisions) with intelligent lifecycle management, including deduplication, confidence decay, and LLM-powered consolidation, ensuring a rich and accurate knowledge base. The system features robust security measures, credential scrubbing, and dual storage for both semantic search and structured queries.
Key Features
01Typed memory with distinct behaviors (event, fact, status, decision)
02Agent isolation and secure API gateway to prevent direct data access
03Dual storage (Qdrant vector database + structured database) for versatile querying
04Automatic credential scrubbing of sensitive data before storage
05LLM-powered memory consolidation, deduplication, and confidence decay
063 GitHub stars
Use Cases
01Share memory and context between AI agents operating on different machines and frameworks
02Provide persistent, structured, and semantically searchable memory for autonomous AI systems
03Obtain concise session briefings on activities from other agents to avoid context loss
