Mnemosyne

Mnemosyne is a memory layer designed for AI agents that runs entirely locally using SQLite, requiring only a single Python dependency and one pip install. It offers sub-millisecond query latency, stores data locally without cloud services or API keys, and integrates with multiple agent frameworks including Cursor, Claude Code, OpenWebUI, and any MCP-compatible client.

The system manages episodic memories, summaries, and knowledge graphs through a browser-based demo or direct Python SDK. It supports negation extraction, instruction parsing, and memory consolidation. The project is open source and available on OpenCollective, with active development tracked on GitHub.