OpenRig is a command-line rig for AI coding agents: you describe a team of Claude Code and Codex seats in a YAML file, and one command boots the whole team inside tmux. It is aimed at developers who already run several coding agents at once and are tired of keeping track of them as a pile of unrelated terminal windows. The project is written in TypeScript, published on npm as @openrig/cli, released under Apache 2.0, and has collected about 1,500 stars and 128 forks since it first appeared.
What it does
The README states the idea in one line: a harness wraps a model, and a rig wraps your harnesses. Claude Code and Codex are each a harness around a model, with their own session, their own prompt loop and their own terminal. OpenRig sits one level above them and treats both as seats in a single system, so that a mixed team of agents is something you configure once rather than assemble by hand every morning.
What you get out of that is a persistent, organized team instead of scattered sessions. You talk to a lead agent about the outcome you want; it can coordinate specialists across teams and bring back results and the decisions that actually need your attention. In the flow the project shows, a second agent checks the work of the first, so review is part of the rig rather than something you remember to ask for. A live dashboard displays every seat with its model, its context and its state, which is the piece that is hardest to reproduce by opening tabs yourself.
The README also recommends a modest starting point: begin with one repository and one useful change, then keep the team's work and context at the same addresses so they stay findable as the project grows.
How it works
The substrate is tmux, which is why the tool is Unix-only and why sessions survive when you close your terminal. Your team definition lives in YAML, so a rig is a file you can read, diff and commit alongside the code it works on, rather than a set of hand-typed commands.
Booting a rig is not a purely read-only act, and the project is upfront about that. Launching one writes provider hooks and workspace trust settings on your machine. The README asks you to read its "what OpenRig changes on your machine" section and back up the relevant files before you run anything — advice worth following, since these are the same configuration files your existing Claude Code and Codex setups depend on.
Getting started
Installation is a global npm package, and the first command is deliberately a rehearsal:
npm install -g @openrig/cli
rig setup --dry-run
The dry run prints a plan, including a check of both the native harnesses and cmux; you review that plan before applying rig setup for real. Requirements are Node.js 20, 22 or 24 plus tmux, on macOS or Linux. On Apple silicon Macs the README points you at Node.js 22 specifically, and links to a release note explaining the compatibility limitation. Native Windows is not supported, and WSL2 has not been tested.
Bun users can install with bun add -g @openrig/cli, but OpenRig still runs on Node.js, so Node 22 has to be present as well. The README warns that Bun may block the package's postinstall script, in which case the Node.js and SQLite check does not run at install time — worth knowing if something misbehaves later and you never saw a warning.
When to use it / when not
This is a good fit if you routinely have more than one agent working, want a lead-and-specialists structure with a built-in checking pass, and want a single screen that tells you what each seat is doing and how much context it has left. It is also a reasonable fit if you want that arrangement written down in a file rather than reconstructed from memory.
It is a poor fit in several concrete cases. If one agent in one terminal covers your work, the rig adds configuration without buying you much. If you are on Windows without WSL2, you are outside the supported set. If you are uneasy about a tool writing provider hooks and trust settings into your environment, read that section of the README first and decide before installing, not after. And as a young project — first commit this April, still pushing changes this month — expect the setup surface to move between versions.
Anyone running Claude Code and Codex side by side on the same codebase should look at OpenRig seriously, because it addresses the specific problem that comes with that setup: not the quality of any single agent, but the absence of any structure holding several of them together. The YAML definition, the tmux persistence and the dashboard are the three pieces that turn a set of sessions into something you can manage, and they are free, open source, and simple enough to try with a dry run before committing to anything.