OpenRig
Get started ↗
MenuClose

Built for persistent populations of agents.

Talk to one agent. Build with a whole team.

OpenRig connects Claude Code, Codex and Pi into persistent teams that run on your own machines. Open source.

One agent in one session can do impressive work. Past that, you run out of rope: it forgets what it learned, and running several agents becomes your full-time job. OpenRig is for everything after that. Bring your ideas to a lead agent: it hands work to specialists, keeps track of it and brings the answers back.

See the team at work ↓Get started →

Claude Code · Codex · Pi · Self-hosted

Work with your team from the tool you already use.

  • Your own Claude Code or Codex: the openrig plugin lets you hand off a task, follow it, answer the team's questions and read the result without leaving your session.
  • Zed or T3 Code: rig acp serve turns a prompt into a task for your team, and the answer comes back as the reply.
  • Claude Desktop and other MCP apps: rig mcp serve lets them see your team, hand out tasks and read results.
  • VS Code and Cursor: the OpenRig extension, in early preview.
Find the tool you already use →

What your team can do

See everything OpenRig does →

One conversation. A whole team behind it.

Coordinate work that takes weeks.

Split a big project across agents that each own a part of it. They coordinate with each other, and each can use its own short-lived helpers for small jobs. The team can hold more than any one agent could keep in view.

Example: you ask for a set of big app features over several weeks. dev-lead gives the screens to dev-ui and the backend to dev-app. Each picks up its own task and uses a couple of helpers to check code and tests. When dev-ui needs to know how the backend connects, it asks dev-app directly and gets the answer: reuse the shared client. dev-lead keeps track and brings the results together for your review.

Illustration: names and output are made up.

How this works →

Reach the right expertise

Ask the specialist.

On a project with months of decisions behind it, another agent may already know the answer. rig send asks that agent directly and brings the answer back, without opening a task for a quick question.

Example: dev-ui asks dev-app how a new feature should connect. dev-app remembers deciding this last month: use the shared client, which now handles reconnecting. dev-ui reuses it.

Illustration: names and output are made up.

How this works →

Keep track of the work

Assign tasks with rig queue.

Some work has to stay visible after the conversation moves on. With rig queue, an agent hands a task to a named teammate, who picks it up. The task stays assigned, with its progress recorded, until it's done.

Example: you ask product-lead to start the next feature. It creates a task for dev-app with the requirements. dev-app is notified, picks it up and reports back, and product-lead sees who owns it.

Illustration: names and output are made up.

How this works →

Explore the team in more detail →See every feature →

OpenRig TUI

See what your whole team is doing.

How your agents are organized, what each is working on, and where they need you, in one terminal view, without checking every agent's terminal.

Click into the demo and try it.Open a pod, pick an agent, or switch to a team's terminals.

On a phone you'll see part of the view. Open it on a wider screen to see every column.

Demo data is made up; nothing runs.

Latest release · OpenRig 0.7.15 · 14 November 2026

A steadiness release, with fewer stops after an update.

When an OpenRig update changes its own Codex hooks, OpenRig now renews their approval itself, unless you've made your own choice about them, so your Codex agents don't stop at Codex's hook review. A slow daemon start now gets three minutes before it's reported pending, and there are fixes for restarts, Slack replies and terminal views, many from people using OpenRig.

Read what's new →

Between mid-September and 0.7.11 we shipped 30 releases and merged about 2,100 pull requests, hundreds of them from contributors outside the OpenRig team, adding more than 40 new features. Two months of OpenRig →

Most of my agents aren't working right now.

Terminal roster on one Mac mini showing running and stopped agent seats.
One Mac mini · September 30, 2026 · 109 agents defined, 41 running, 13 working.View full-size roster ↗

I start agents as I need them and leave them on. They build up context, and later I send work to whichever one already knows that part of the project.

That's a persistent population: agents you start once and keep. Each one builds up context in its part of the project, so the next piece of work goes to the agent that already knows it. Start with a small team, such as Workshop's four agents, and keep it on.

Next steps

From the notebook

From the blog

Read the blog →
Notes / 2026-11-12

Two months of OpenRig: new features and improvements

What's new and what got better in OpenRig from 0.5.15 to 0.7.11, by feature.

Notes / 2026-10-24

This week's Node.js, PHP and GitLab advisories, and OpenRig

What the Node.js, PHP and GitLab security releases of October 21 mean for OpenRig. Only the Node.js one applies: update Node.

Notes / 2026-10-22

Some roads lead to other minds

The agent that solved a problem last month can still be useful to the agent working on it today. Here is how I connect project documentation, specialist agents and saved history.