Repolode
Submit
NevaMind-AI

NevaMind-AI/memU

A lightweight memory system that gives coding agents and desktop AI hosts a shared, cross-session memory stored as a wiki.

View on GitHubWebsite
Stars
14,556
Forks
1,090
Open issues
134
Last commit
11 hours ago

What is it?

memU is a memory layer for AI agents. It reads session logs from desktop agents such as Codex, Claude Code, Cursor, OpenClaw and Hermes Agent, distills them into memory and reusable skill Markdown files, and retrieves relevant memory before a future task. Memory is shared across hosts through one configured backend, so what one agent learns can be used by another.

The system runs as a sidecar with one CLI adapter per host, plus a generic memu-agent command that detects other agents. A scheduled bridging task slices new sessions into job files, the agent writes the memory or skill files, and a commit step stores them with embeddings. Storage can be local SQLite, Postgres with pgvector, or MemU Cloud, selected through a config file and environment variables.

The core memory logic is described as roughly 500 lines of Python, and the project is licensed under Apache-2.0. Installation is available through pip, npx or uvx, and a doctor command checks that the retrieval path works.

Key features

  • 01Captures session history from desktop coding agents through per-host CLI adapters.
  • 02Distills sessions into reusable Markdown skills that an agent can patch or create.
  • 03Shares one memory backend across Codex, Claude Code, Cursor, OpenClaw and other hosts.
  • 04Supports local SQLite, Postgres with pgvector, or MemU Cloud storage.
  • 05Provides a doctor command to verify the retrieval path end to end.

Use cases

  • Keeping project context and conventions available across Claude Code and Cursor sessions.
  • Reusing workflows an agent learned earlier when a similar task comes up again.
  • Giving several desktop AI tools one shared memory store on the same machine.
  • Adding persistent memory to an application's own conversations via the memorize interface.

Quick start

  1. 1

    Get an API key from memu.so

  2. 2

    Install the CLI with pip

    pip install memu-cli
  3. 3

    Retrieve relevant memory for your host agent (e.g. Codex)

    memu-codex retrieve "What should I remember about this project?"

Things to know

  • Self-hosted use requires your own embedding provider API key.
  • Chat modes of ChatGPT and Claude are not supported; only specific work and desktop modes work.
  • Agent support varies by operating system, and Linux covers fewer hosts than macOS or Windows.

Topics

#agent-memory#llm#mcp#skills#vector-search#sqlite#postgres#claude-skills#harness#loop-engineering

FAQ

What is memU?
A lightweight memory system that gives coding agents and desktop AI hosts a shared, cross-session memory stored as a wiki. memU is a memory layer for AI agents. It reads session logs from desktop agents such as Codex, Claude Code, Cursor, OpenClaw and Hermes Agent, distills them into memory and reusable skill Markdown files, and retrieves relevant memory before a future task. Memory is shared across hosts through one configured backend, so what one agent learns can be used by another.
Is memU free to use? What license does it have?
memU is open source under the Other license. Check the license terms before using it in commercial products.
Is memU still maintained?
Actively maintained. The last commit to memU was on Oct 20, 2026.
What can I use memU for?
Keeping project context and conventions available across Claude Code and Cursor sessions. Reusing workflows an agent learned earlier when a similar task comes up again. Giving several desktop AI tools one shared memory store on the same machine. Adding persistent memory to an application's own conversations via the memorize interface.
How do I install memU?
Get an API key from memu.so Install the CLI with pip: pip install memu-cli Retrieve relevant memory for your host agent (e.g. Codex): memu-codex retrieve "What should I remember about this project?"
Where can I get memU?
The source code and installation instructions are on GitHub: https://github.com/NevaMind-AI/memU

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects