Agent Magnet
@helinakdogan
Self-learning memory for AI tools. Remembers user preferences and context across Claude, Cursor, and Codex — with multi-parameter forgetting and cross-tool identity.
Overview
What is Agent Magnet?
Agent Magnet is an open-source memory infrastructure for AI tools. It automatically learns user preferences, corrections, and patterns and persists that memory across sessions and across different AI tools like Claude, Cursor, and Codex.
How to use Agent Magnet?
Install with pip install agent-magnet and run agent-magnet init to configure automatically. The init command detects your installed AI tools (Claude Code, Cursor, Claude Desktop) and writes the correct configuration files. No Redis or external services are required by default — local SQLite storage works out of the box.
Key features of Agent Magnet
- Cross-tool memory across Claude, Cursor, and Codex
- Self-learning from corrections and patterns automatically
- Multi-parameter forgetting with permanent, contextual, and transient decay
- Team memory pool where new members never start from zero
- On-premise Docker deployment, no data leaves your infrastructure
Use cases of Agent Magnet
- Keeping coding style and project context across AI tool switches
- Building AI products where users don’t repeat themselves every session
- Team onboarding — new developers inherit the team’s accumulated knowledge
- Enterprise AI deployments requiring data sovereignty
FAQ from Agent Magnet
Is Agent Magnet free to use?
Yes, the core SDK and MCP server are open-source and free. A hosted proxy option is available at agentmagnet.app.
Does it require Redis or external services?
No. By default it uses local SQLite storage. Redis, Qdrant, and Neo4j are optional for teams and enterprise.
How do I share memory across Claude Code and Cursor?
Set the same MAGNET_USER_ID in both tools’ configurations. Agent Magnet uses this to link memory across tools.
