OpenRig is a free, open-source (Apache 2.0) multi-agent harness for organizing Claude Code, Codex, and Pi into a persistent team. Users can communicate ideas and feedback with just one lead agent, which delegates work to the various specialist agents, tracks their progress, and brings answers back to the user. It is officially positioned as a persistent coding agent team that can collaborate on projects spanning weeks or months, and it supports self-hosting.
[Main Capabilities]
- Persistent team: lets Claude Code, Codex, and Pi work together as a long-lived team rather than a one-off session.
- Lead agent delegation mechanism: users submit ideas and feedback to the lead agent, which assigns tasks to named specialists (such as dev-lead@factory, dev-ui@factory, dev-app@factory in the examples).
- Its own task queue: assign work to named teammates via rig queue; teammates read and claim tasks, and the task remains owned by that teammate with progress recorded, making follow-up easier.
- Consult specialists directly: when a project has accumulated a large number of historical decisions, you can use rig send to ask a specialist agent directly and bring the answer back into the current work, without having to create a new task for a simple question.
- Terminal UI overview: the OpenRig TUI presents the team's organizational structure, the work currently being handled, and the places that need the user's attention in one place, without having to check each agent's terminal one by one.
- Build your own agent population: officially described as "Build your own agent population," allowing you to build your own agent team.
- Cross-framework collaboration: the official article mentions that Claude and Codex each have their strengths, and that extending across harnesses is more important than using sub-agents within a single harness.
[How to Use] The official getting-started steps include: run Claude Code or Codex in the project folder, then have Claude Code set up OpenRig; afterward, tell OpenRig's operator what you want to build, and it will recommend a team and start it after the user confirms. Prerequisites: Node 22 or 24, tmux, and a signed-in (signed in) Claude Code or Codex; on Apple silicon, Node 22 is recommended. The team is defined by YAML (mentioned in the description); the specific configuration fields and complete installation commands are not yet provided.
[Applicable Scenarios] Suitable for splitting a large project among agents that excel at different parts, and letting each agent maintain its own responsibilities, coordinate with teammates, and when necessary use its own temporary sub-agents to handle smaller tasks. Official examples include: assigning interface work and backend work to different persistent specialists; specialists each using temporary native sub-agents to perform limited checks and test coverage issues; specialists asking each other about backend connection methods and reusing existing decisions. It is also suitable for long-term maintenance projects, reusing decisions from months ago by directly asking the specialist who maintains that part.
[Limitations and Notes] The TUI interaction examples and agent terminals on the official page are demonstrations using fictional data; the spawn_agent and Claude Agent calls mentioned in the text are illustrative examples, and the returned results are all fictional, and the helper has no OpenRig address or queue ownership. The applications and shared clients mentioned in the text are all fictional. OpenRig's own charging model, pricing details beyond the Apache 2.0 license and "free, open-source, self-hosted," the range of supported platforms, and so on are not yet provided.
