AgentDB

A specialized vector database designed for AI agents to manage long-term memory and cross-session state.

AgentDB solves the persistent memory gap for LLM agents by providing a dedicated storage layer for context and history. Built by Matt Shumer, it enables developers to save, retrieve, and query agent experiences using simple Python commands (like .remember and .recall). The system uses semantic search to surface relevant past interactions, ensuring agents maintain continuity across thousands of distinct tasks without manual context window management.

https://github.com/mshumer/AgentDB