NevaMind-AI/memU ? reverse-engineered prompt

Build me a Python tool called memU that gives coding agents a shared personal memory across sessions, agents, and devices.

I want it to watch agent session history, pull out useful things the agent learned, save them as simple Markdown wiki entries, and then bring the right memory back later when a similar task comes up. It should work as a sidecar for desktop coding agents, with a setup flow that patches the right instruction file and adds a command for retrieve before answering. Please include support for the main agents mentioned in the docs like Claude Code, Cursor, Codex, OpenClaw, and a generic fallback for other agents it can detect.

Keep the core memory logic small and easy to inspect, and make it possible to run either with a cloud API key or self hosted with a local embedding provider. If you need to check current integration details, look them up online. Also make sure the install and uninstall experience is clean and safe, with memory kept unless the user explicitly asks to erase it.