memU is a lightweight memory system for language model agents that persists knowledge across sessions, agents, and devices by storing it as a wiki.
The tool addresses the problem of agents losing context and reusable knowledge between separate interactions. It works by automatically capturing session history, distilling reusable skills from agent interactions, and storing them in a retrievable format that can be accessed across different agents and devices. The core memory logic is implemented in approximately 500 lines of code, making it small enough for users to inspect and modify. The system supports both memorization—capturing useful knowledge through background tasks—and retrieval, bringing relevant stored memory into future tasks.
Developers should choose this tool if they work with multiple coding agents across different sessions and want to build up a persistent knowledge base without manual effort. It integrates with agents including ChatGPT, Claude Code, Cursor, OpenClaw, Hermes, and WorkBuddy across macOS, Windows, and Linux. The tool is free and unlimited, with setup involving an API key from the project's website. Support varies by agent and operating system; for example, ChatGPT's chat mode does not currently support memorization or retrieval, while Claude Code works on desktop and CLI but may require retrying setup with specific models.
Almost all open issues are raised by outside users rather than the core team, indicating a substantial base of adopters reporting real-world use. Maintainers typically respond to new issues and pull requests within a day. Work in the issue tracker is dominated by bug reports, enhancement requests, and items related to a community challenge.
