OpenRig is the Terraform for Coding Agents (Launch Demo)

Video

Your coding setup might look like a pile of terminals. A bunch of Claude Codes, maybe Codex. A bunch of tabs, no IDE. Right now it feels like it's all about how many of these you can juggle at once, and whether you can keep them going.

For a while, my personal max was about three or four in parallel. If you've tried doing that for a full day, you know exactly how hard it is.

So I ended up making a tool that lets me juggle way more than that. It's called OpenRig. It's open source, and it's a CLI for your coding agents. Once it's set up, your agents can message each other and control each other's terminals. That lets them coordinate, so you don't have to babysit them anymore. I mean, you do a little bit, but way less than normal.

You talk to one agent

Most of the time I only talk to one agent, the orchestrator, and it handles the rest.

Its job is to keep the other agents on task. If one of them does a bunch of work and then stops, even though it was supposed to keep going, the orchestrator nudges it and tells it to keep going. If one gets stuck at a permissions prompt, the orchestrator can just click accept.

Update, September 2026: OpenRig now refuses to answer another agent's permission prompt by default. You can still do it, but only as a deliberate override (rig send --dangerously-interact, with a reason), and it's recorded.

That's what lets the whole rig keep working pretty much indefinitely. It's really up to you how much work you give it.

Your orchestrator should never go down

That makes the orchestrator the one agent you can't afford to lose.

I really don't like resetting my orchestrators, because they hold all the context of what's happening in the rig and the project. It's common for me to have them running non-stop for weeks at a time.

But any agent can get reset or compacted. So how do you keep one going for weeks?

Run two of them

You run more than one. In my topology I have two orchestrators, one Claude and one Codex. They work kind of like a high-availability cluster. If one of them gets reset or compacted, the other one jumps in. When the first one comes back, the other helps it restore its mental model, and they just keep going.

Steer it from anywhere

I usually talk directly to the Claude orchestrator, because Claude has remote control and a great mobile app. That means I can keep managing the entire rig from anywhere.

Getting started

The best way to use OpenRig is to just ask your agent to do everything. That's what I do 99% of the time. The CLI is designed for agents, so yours will figure it out.

OpenRig is free and open source. Since these are just regular Claude Code and Codex sessions, you can use your regular subscription, and you don't pay API costs. If you want to try it, point your agent at openrig.dev. The video shows the rest, including how I define every agent in a single file.