Memory Bank

About

Memory Bank is a production-ready Model Context Protocol (MCP) server designed to equip AI agents with a robust, vector-native memory system. Built upon the Protocol-Lattice Go Agent Framework, it offers persistent, searchable, and shareable memory capabilities, ensuring flexible data storage across various database backends like PostgreSQL, Qdrant, and MongoDB. It supports both short-term contextual buffering and long-term semantic retrieval, complete with dynamic embedding capabilities and shared memory 'spaces' with fine-grained access control, enhancing the intelligence and collaborative potential of AI systems.

Key Features

• Shared memory 'spaces' with access control (ACLs) and time-to-live (TTL)
• Comprehensive set of MCP tools for memory manipulation and diagnostics
• 13 GitHub stars
• Dynamic embeddings configurable via environment variables (OpenAI, Gemini, local models)
• Rich memory management with short-term buffer, long-term persistence, and contextual retrieval
• Multiple vector store backends (PostgreSQL, Qdrant, MongoDB, In-Memory)

Use Cases

• Enabling collaborative knowledge bases and shared memory contexts for multi-agent AI systems
• Storing and retrieving semantic information to enhance decision-making and learning in AI applications
• Managing conversational memory for AI chatbots and virtual assistants across sessions