Agent Magnet

Gives an assistant a persistent behavioral memory of your preferences, decisions, and project context across sessions and tools.

What you can do

Agent Magnet is a memory layer that records decisions, conventions, watch-outs, failed attempts, goals, preferences, and actions as you work, without requiring explicit "remember this" prompts. It sorts these into one memory store that can be shared across Claude, Cursor, and Codex, and lets you forget or mark items done so the memory does not pile up. Storage is local by default, with an optional Redis-backed shared mode.

When to use it

Worth adding if you repeatedly re-explain the same project conventions, decisions, or preferences to an assistant in new sessions, or if you want one memory shared across several AI coding tools. Less useful for one-off, single-session tasks.

Requirements

Runs locally as a Python package (pip install agent-magnet) over stdio. Requires MAGNET_USER_ID as the memory key. Optional: MAGNET_LOCAL_MODE for local SQLite storage, MAGNET_PROJECT_ID for namespacing, MAGNET_REDIS_URL for shared persistent storage, and MAGNET_OPENAI_KEY for the reflector and classifier models. Team memory requires a paid key from the vendor.