Every project. Every agent. Every monitor.
mAgent is a Claude Code & Codex agent manager: it opens each of your projects in its own terminal, launches your AI coding agent inside it, resumes the last session, and auto-tiles every window into a grid across all your screens. One command does it all.
Not the MAgent multi-agent reinforcement-learning library: mAgent is a developer CLI for running AI coding agents.
Procrastination dies when your desk sets itself back up.
The hardest part of deep work is re-entry — reopening projects, hunting for windows, remembering where every conversation left off. mAgent restores the whole workspace exactly as it was, and resumes your latest Claude and Codex conversations mid-thought. You don’t start over. You just keep going.
A whole fleet of agents, run like one workspace.
Your fleet, most-urgent first.
One live table of every agent session. States come from your agents’ own lifecycle hooks — working, needs-input, done, error. Press a row number to jump straight to that window (Windows).
Your whole deck, from anywhere.
On Windows, sessions live in psmux, so they survive you walking away. SSH in from any client and pick a project off the session list — or pair Happy for end-to-end-encrypted control of your Claude and Codex sessions from a browser or phone. All of it over Tailscale, never the open LAN.
Runs everywhere. Fullest on Windows.
Launching, tiling, and the mobile/notification plumbing run on all three OSes. A few power-user features lean on Win32 primitives with no cross-platform equivalent yet — every cell here is derived from the real capability probes, not aspiration.
Questions, answered.
What is mAgent?
mAgent is a free, open-source CLI that opens each of your projects in its own terminal window, launches your AI coding agent inside it, and auto-tiles every window into a grid across all of your monitors. One command takes you from an empty desktop to a fully tiled, resumed AI agent workspace.
Which AI coding agents does it support?
Claude Code, OpenAI Codex CLI, and Cursor CLI (cursor-agent) out of the box, alongside editors like VS Code and Cursor. Any tool that runs from your PATH can be launched per project, so mixed fleets — some projects on Claude, some on Codex — work naturally. Live agent state and automatic conversation resume are wired for Claude Code and Codex.
How do I run multiple Claude Code sessions at the same time?
Run magent --go. Each project gets its own terminal window with Claude Code already running, tiled into a grid across your monitors. Sessions resume with claude --continue automatically, and the magent watch table shows the live state of every session — working, needs-input, done, or error — so you always know which window needs you.
Does mAgent resume my Claude Code conversations after a reboot?
Yes. The default Claude Code command is claude --continue, so magent --go reopens every project in its own terminal and picks up the most recent conversation in each folder. A project with no stored conversation starts a fresh claude instead of failing, and a project with several windows resumes a different conversation in each, by session ID. How to resume a Claude Code session after a restart
Is mAgent free and open source?
Yes. mAgent is open source under the AGPL-3.0 license and free to use. Install it with pip install magent-multi-ai-agents-manager or uv tool install magent-multi-ai-agents-manager; the source lives on GitHub under DevinoSolutions/magent-multi-ai-agents-manager.
Which platforms does it run on?
Windows, macOS, and Linux, with Python 3.10 or newer. A few power features — like psmux persistent sessions and desktop toast notifications — are Windows-only; the capability matrix on this page lists exactly what each platform supports.
Can I monitor and control my agents remotely?
Yes. On Windows, sessions live in psmux, so they survive you walking away — SSH in over Tailscale from any client and re-attach. On every platform you can pair it with Happy for end-to-end-encrypted control of your Claude and Codex sessions from a browser or phone, and upload screenshots from your phone; on Windows an uploaded screenshot drops straight into the focused session.
What do I install, and what do I run?
Install it from PyPI with pip install magent-multi-ai-agents-manager (or uv tool install magent-multi-ai-agents-manager). The command is magent — run magent once to scan your history and write a config, then magent --go to launch and tile every project.
Can an AI agent set mAgent up for me?
Yes. Copy the "Onboard your agent to mAgent" sentence from the pill on the home page or the /agent-setup page and paste it into Claude Code, Codex, Cursor, or any AI coding agent. It fetches /agent-setup/prompt.md and runs the install, config, and (for Claude Code and Codex) lifecycle-hook steps itself — no credentials needed.
Is mAgent the same as the MAgent reinforcement-learning library?
No. mAgent (magent.now) is an open-source command-line tool for developers that opens, resumes and tiles AI coding agents such as Claude Code and Codex. It has nothing to do with MAgent, the many-agent reinforcement-learning platform; the two only share a name.
Do I need an API key or account to use mAgent?
No. mAgent has no API key, no OAuth flow, and no MCP server — it is a local CLI that launches terminals and agents you already have installed. It never talks to a mAgent-run service.
still have questions? ask in our Discord →
Two minutes to a tiled workspace.
Install, then run magent once to scan and configure. Run it again to launch everything.
pip install magent-multi-ai-agents-manager
Or skip the menu: magent --go launches and tiles, -g <group> launches one group, doctor diagnoses your environment.
step-by-step install guide