MEX
Team project memory for engineers and agents: architecture, decisions and handoffs live in the repo and travel through Git
Install
npx mex-agent@0.8.2 setupOpens setup in the local browser Hub. Add --cli for terminal setup.
This is third-party code. Review the repository files before installing.
What it does
MEX keeps a project wiki next to the code: architecture, decisions, conventions and patterns linked to code. Proposed knowledge changes land in an inbox and are accepted explicitly, and the mex-relay skill drafts a handoff for a teammate with progress, decisions, blockers and next steps. People browse and review memory in a local web Hub; agents read and update it through project instructions and the CLI. Everything is Markdown shared by ordinary commits, with no hosted service or account.
Who it is for. For teams where several engineers with agents pick up each other's work.
Good fit when
- Knowledge from agent sessions should persist and reach teammates
- You need to hand off unfinished work to another engineer with context
- You want architecture decisions in the repo with reviewed changes
Not a fit when
- The project is not in Git
- A simple note for one agent is enough; use CLAUDE.md or AGENTS.md
Example request
Draft a handoff of this task for Sam: what changed, which tests ran and what remainsLimitations
Requires Node.js 22.5 or later with SQLite FTS5 support and a Git repository. Memory population during setup is done by Claude Code or Codex CLI, which have their own account and network requirements. MEX never pushes or pulls; sharing happens only through your commits.
How to disable. Remove MEX files and pointers to them from agent instructions, then commit.
Security check
- Runs a local web server and indexes project code
- Changes agent instructions and adds files to the repo
README in short
The README explains what knowledge a team keeps in MEX and walks through a handoff between two engineers. It describes the local Hub with wiki, code, inbox, relays and activity. Setup runs via npx mex-agent setup in the browser or with --cli in a terminal, with an optional contact form via Web3Forms at the end. MIT licensed.
FAQ
Do I need a server or account?
No, memory lives in the repo and each teammate has their own local indexes and Hub.
Does a relay notify the teammate?
No, publishing only writes files. The teammate sees it after you push and they pull.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent