OpenRig
Status check 2026-11-13 — active
Verdict: There were 14 releases between v0.7.0 (2026-10-23) and v0.7.13 (2026-11-12), and several commits landed on main on 2026-11-13. Last release: v0.7.13 (2026-11-12), also published on npm as @openrig/cli · last commit: 2026-11-13 · licence: Apache-2.0 Repo: https://github.com/mvschwarz/openrig Changed since this note: The note says the current release at ingest was v0.6.8, but v0.6.8 is from 2026-10-09 and v0.7.6 was already out by the note date. Today’s version is v0.7.13. v0.7.0 (2026-10-23) added a human gateway with quiet hours and modes (asleep, out of office), rig names usable in place of ids…
🚀 Description
mvschwarz/openrig — “A harness wraps a model. A rig wraps your harnesses.” You define an agent team as a RigSpec in YAML (pods, edges, continuity policies), boot it with rig up, and talk to a lead agent about the outcome while it coordinates specialists. Claude Code and Codex can sit in the same rig; Pi and Oh My Pi attach through RPC runners. Open source, local-only, built on tmux. Current release at ingest time: v0.6.8.
It tackles terminal sprawl: which sessions are running, how they relate, how to recover after a reboot.
🧩 Features
- Define / boot / see: RigSpec topologies,
rig up, a TUI with topology table and graph, Specs, Projects, Terminals, Feed - Adopt existing Claude Code and Codex sessions found in tmux
- Snapshot and restore:
rig down --snapshot, thenrig up <name> - Communicate:
rig send,rig broadcast,rig chatroom, plus an owned-work queue (rig queue) - Grow and shrink a running topology (
rig grow,rig shrink) - MCP server so agents manage their own topology (
rig_up,rig_ps,rig_send, …) - Built-in teams:
starter(Claude Code builder + Codex reviewer for one change),workshop(lead, builder, QA, reviewer for one repo),factory(seven agents: lead, advisor, build, QA, design, two independent reviewers); specialist teams for code review, research and PM - Architecture: CLI / TUI / MCP → Hono HTTP daemon → domain services → SQLite + tmux + runtime adapters
- Install: Node.js 22 or 24 and tmux, macOS or Linux — no native Windows
⚠️ What it changes on your machine
Worth reading before installing — rig setup and the daemon write trust settings and executable hooks:
- edits
~/.tmux.conf,~/.claude.jsonand~/.codex/config.toml(workspace trust, hooks, activity relays) - seeds skills into
~/.claude/skillsand~/.agents/skills - managed seats launch Claude with
--permission-mode acceptEditsand Codex with-s workspace-write; theworkshopbundle bypasses permission prompts entirely - the documentation itself says to back up provider files first; hook and trust entries can be replaced
Reasoning for
The most complete answer so far to “how do I run Claude Code and Codex as one persistent team instead of loose terminals”. Relevant to Swarm Research — Orchestrating Coding Agents and Agentic Systems. Not usable on this Windows box without WSL, and the broad config rewriting makes me want a throwaway machine for any trial.
Alternatives considered
- oh-my-pi — a single agent with LSP and a debugger, not a team layer (OpenRig can host it)
- Claude Peers MCP (retired) — lightweight Claude-to-Claude messaging without topology management; Claude Code now has this built in
- Hermes Agent — agent gateway rather than a multi-harness team manager
- Harness Engineering — the concept; OpenRig is one implementation of “harness of harnesses”
🔗 Links
- Repo: https://github.com/mvschwarz/openrig
- Getting started: https://github.com/mvschwarz/openrig/blob/main/docs/reference/getting-started.md
- Site: https://openrig.dev
Template: tool