ai-ready label; Orbi Cloud claims it, writes the code in an isolated workspace, runs your test suite, opens a pull request, reviews that pull request in a separate session, repairs what the review finds, and merges the exact reviewed commit. The Issues, the pull requests and the release evidence all stay in your repository. There is no second workspace to adopt and no new password.
Orbi Cloud runs the same delivery engine as self-hosted Orbi, which is open source. The difference is who operates the runner: with Cloud, Orbi does.
The whole path, end to end
Everything below the dashed line is automatic. Your part is the four steps of setup, then one label per piece of work — and the decision to cut a release. The two dotted arrows are the only two switches you throw: theai-ready label, and the version number. Orbi never throws either one for you.
What you actually do
Sign in with GitHub
One OAuth click. Your GitHub account is your Orbi identity — there is no separate credential.
Install the GitHub App
You choose which repositories the runner may work in, and you can revoke it from GitHub at any time.
Connect one repository
Pick the repository and its base branch. Orbi provisions a runner for it, usually in about a minute.
Label an Issue ai-ready
That label is the execution switch. Everything after it is automatic until a pull request is waiting for you.
The delivery loop
The review is a separate session that did not write the code. It repairs the findings it can and re-runs the tests; only a clean verdict reaches the merge gate. When automatic recovery is not safe, the Issue lands onai-blocked with a comment saying why — that state is a deliberate stop, waiting for a human decision, not a crash.
Read the delivery lifecycle for what each label means and what moves an Issue between them.
What it costs
Your first 3 merged deliveries are free — no card and no subscription. Failed deliveries do not use up that allowance. After it, Orbi Cloud is US$79 per month, which includes 300,000,000 tokens of model usage per month. There is no per-token overage billing: when the monthly allowance runs out, new deliveries pause until the next calendar month. See limits and quotas for how the allowance is counted and what happens at the cap.What Orbi Cloud does not do
It does not discover repositories on its own — it only ever works in the one repository you connect. It does not start a release on its own; you open the release Issue and apply theai-release label, which only a human can set. It does not push to your protected branches outside of the reviewed merge, and it does not touch a delivery that is already running when the platform is upgraded.
Read security for the full boundary list, including where your model key is stored and what the runner can reach.
Orbi Cloud is in Private Beta. One repository can be connected at a time, and the walkthrough here reflects the product as it currently ships — not a roadmap.
Where to go next
If you want the shortest path to a merged pull request, start with the quickstart. If something is already wrong, go to troubleshooting.The engine’s own documentation
These pages cover Orbi Cloud — the hosted product. The delivery engine underneath is open source and documented separately at docs.orbi.build. Go there when you want the mechanism rather than the product:The delivery workflow in depth
The full label state machine, the review and repair loop, steering, and the release state machine — the engine’s own reference for everything Cloud runs on your behalf.
The engine's security boundaries
What the engine will and will not do: what it never pushes to, what it never merges, and why.
Run it yourself instead
Self-hosting is free and stays free. Same engine, you operate the runner.