AQ Crews
A team of agents,reviewed by another
Pick the Ship, Review or Fix crew like you pick a model. A Lead plans, you approve, builders work in their own worktrees, and Auto prefers another vendor and account for the review.
An AQ crew is a named team of AI agents for a kind of work. The Lead reads the task, proposes a plan, and delegates each step to a role: Build, Test, Research or Review. Each role runs on the agent and account your organization chose for it, and Auto looks for a reviewer on a different vendor and account from the builder. The crew summary shows whether that review is independent.
Three crews, ready to use
Ship
Build a feature
Plan, build in slices, an independent review, tests, then a pull request.
Review
Review a change
A read-only review of the branch, bound to the commit it reviewed, with a preference for another vendor.
Fix
Fix a bug
Reproduce the bug, fix it, review it independently, and test it.
Copy a crew to change its playbook, or write your own crews and roles in Settings. You can also just say it: "run this as the Ship crew" works from any work item.
You approve the plan
With the plan gate on, the Lead proposes its steps and waits. Approve, edit the plan, or tell it to do the work in one thread instead. No role that can edit code starts before you approve. Planning still uses the Lead's subscription or AQ credits.
Review that is actually independent
A model reviewing its own work tends to agree with itself. When Auto chooses the reviewer, it prefers another vendor and another account than the builder's, and the crew summary says what that bought you, for example "Review on Codex caught 2 issues before merge". If no independent subscription is available, the summary says so; AQ credits are used for a reviewer only when you allow them in Crews settings. Security reviews can run on Hinoki, a model built for offensive security. Background: the self-review problem.
Auto suggests a crew when it helps
On a thread routed by Auto, a turn that rates as a hard feature, refactor, migration or bug offers to run as a crew once. Keep one thread, or say yes and the Lead takes over with your message.
Crews run on the agents and subscriptions you already have.
Frequently asked questions
What is a crew of coding agents?
A crew is a team of AI agents with roles: a Lead that plans and delegates, and builders, testers, researchers and reviewers that each run as their own thread, on the agent chosen for that role.
Why does the reviewer use a different vendor?
Auto prefers a different vendor and account to bring another perspective to the review. It depends on your connected subscriptions and review credit settings; the crew summary tells you whether the review was independent.
Do crews cost more than a single agent?
Each role runs on the subscription or AQ model picked for it. With your own subscriptions the roles use your plans' included usage; AQ models are billed at their per-token price. The plan gate means nothing is built before you approve.
Can I make my own crew?
Yes. Copy Ship, Review or Fix and edit its playbook, or create a crew and custom roles from scratch in Settings.