memU
A lightweight, agent-driven memory system that gives coding assistants and other AI agents one shared memory wiki across sessions, agents and devices, and distills reusable skills from their history.
memU is a lightweight, agent-driven memory system that gives a user one shared memory layer, stored as a wiki, across sessions, agents and devices. It automatically distills reusable skills from an agent's own history, and its core memory logic is small enough, at around 500 lines, to read and adapt directly.
It works with a range of hosts, including Codex, Claude Code, Cursor, OpenClaw, Hermes and others, and offers two core operations: memorizing, which captures useful session knowledge through a scheduled background task and turns it into reusable memory, and retrieving, which brings relevant memory into a later task. Support for each combination of host and operating system is tracked and tested individually, and some combinations carry noted limitations, such as needing a specific model or a particular memU version for retrieval to work correctly.
It is free to use across devices, and is set up by sending an agent installation instructions along with an API key obtained from the project's site.