dalao.ai

OpenRig

Run teams of Claude Code, Codex and Pi agentsOpen source

About OpenRig

OpenRig is open source multi-agent orchestration summed up in one line: a harness wraps a model, and a rig wraps your harnesses. Define a team of Claude Code, Codex and Pi agents in YAML and boot it with one command. It turns a pile of terminal sessions into a persistent team with roles, where a lead agent coordinates specialists and brings back results and decisions that need you. It is Apache-2.0 licensed, and as of 2026-10-30 the latest release is 0.7.5, published 2026-10-29.

EditorialUpdated 4 min read

What problem it solves

Run a few coding agents at once and you soon have terminal sprawl: which sessions are running, how they relate and how to bring them back after a reboot all live in your head. The README says OpenRig manages not the agents themselves but the system they form together.

Teams are declared in YAML (RigSpec) with pods, members, edges and continuity policies. rig up starts the tmux sessions, harnesses and startup files and checks readiness; rig down --snapshot saves the topology for restoring by name later. Every agent runs in a tmux session you can attach to directly. The author describes it as the system behind his AI civilization experiments.

A harness wraps a model. A rig wraps your harnesses.

OpenRig README

Core capabilities

Define and boot. Describe the team in YAML, start it with rig up, and grow or shrink it while it runs.

See. A terminal UI with topology tables and graphs shows each seat's runtime, model, context use and state, plus projects, feeds and instance health, with herdr or cmux laying agent terminals side by side.

Adopt. Discover Claude Code and Codex sessions already running in tmux and bring them into a managed rig.

Talk. rig send, rig broadcast and rig chatroom between agents, MCP tools so agents can manage their own topology, and a typing guard that holds automatic messages for seats you type in by hand.

Runtimes. Native Claude Code and Codex sessions, Pi and Oh My Pi through RPC runners, and plain terminal nodes. An experimental Slack link works through an app you create yourself.

Ready-made teams. starter (a builder and a reviewer for one bounded change), workshop (lead, builder, QA and reviewer for ongoing work in one repository) and factory (seven agents including an advisor, design and two independent reviewers), plus code review, research and product specialist teams and a HashiCorp Vault run by an agent.

Who uses it

As of 2026-10-30 the repository, created 2026-04-01, has about 8,302 stars and 685 forks, with about 76 contributors with GitHub accounts (as of 2026-10-29). Releases come fast: 0.7.0 on 2026-10-23, then 0.7.1 and 0.7.2 on 2026-10-25, 0.7.3 and 0.7.4 on 2026-10-27, and 0.7.5 on 2026-10-29: six releases in seven days.

Named teams or companies using it: not found.

How it compares

The README compares itself with Claude Managed Agents: OpenRig is open source and self-hosted, puts Claude Code and Codex in the same team, and runs on your own machine, while model usage is still billed by your providers.

Compared with opening several agents in tmux by hand, it adds declarative topology, snapshots, a single view of everything and message channels between agents. Third-party comparisons of results or cost against other orchestration tools: not found.

Who it's for / Not for

Who it's for

Developers already using Claude Code and Codex who want them to split work into building, review and QA.

People at home in tmux who want agent teams that survive a reboot.

Anyone who wants to manage multi-agent teams as versionable config.

Not for

Single-agent users with simple tasks.

Native Windows users; macOS, Linux or WSL2 is required.

Anyone who does not want global config changes, or who needs tight limits on agent permissions; see the risks below.

Getting started

Prerequisites: Node.js 22 or 24 (22 on Apple silicon), tmux, macOS or Linux (WSL2 on Windows), and Claude Code or Codex signed in. One provider is enough.

  1. The suggested route is to add OpenRig's rigs skill to Claude Code or Codex with npx skills add, then type /rigs (or $rigs in Codex) and describe what you want. The agent explains what it will run and change, installs OpenRig and opens the view.
  2. Or install it yourself with npm install -g @openrig/cli, preview with rig setup --dry-run, then run rig setup. The official install script also supports --dry-run.
  3. Talk to the kernel's operator agent, which asks what you want to build, recommends one of the three teams and fits it to your providers.

Pricing & license

OpenRig is free and Apache-2.0 licensed (as of 2026-10-29). It runs on your own machine, and model costs come from your Claude Code or Codex accounts.

Risks & limits

It changes your machine. The README lists what it writes: a global CLI install; a block in ~/.tmux.conf, possibly installing tmux, herdr or even a missing Claude Code or Codex; hooks and trust settings in Codex config; workspace trust, activity hooks and a status line in Claude Code config; and managed guidance in a project's AGENTS.md or CLAUDE.md. It says it cannot guarantee to preserve or roll back those files, so back them up first.

Broad default permissions. The kernel agent gets more access than project teams by default: full access with no approval prompts on Codex, and file tools, web fetches and operational commands, including reads under your home folder, without asking on Claude Code. The workshop team ships with broad access and a non-interruptive mode that bypasses permission prompts. Permission policies and per-seat settings can tighten this.

Moving fast. Still 0.x with several releases a week; the older React web UI gets only best-effort maintenance.

Our verdict

OpenRig turns running a handful of agents into a system with topology, snapshots and message channels, and it is unusually candid about every change it makes to your machine and its default permissions. If you already use Claude Code and Codex and want them to work as a team, it is worth a try.

Back up ~/.claude, ~/.codex and your tmux config first, read the rig setup --dry-run output, start with the starter team, and tighten default permissions as needed.

Sources

Official

  1. OpenRig READMEGitHub · Accessed Oct 30, 2026
  2. mvschwarz/openrig on GitHubGitHub · Accessed Oct 30, 2026
  3. OpenRig releasesGitHub · Accessed Oct 30, 2026
  4. OpenRig contributorsGitHub · Accessed Oct 30, 2026
  5. OpenRig LICENSE (Apache-2.0)GitHub · Accessed Oct 30, 2026

Updates & reading