Ask
Type a question or describe what you want built, and send it. That is the whole interface. You do not have to choose a model, write a prompt in a particular shape, or decide how hard the question is — the room picks a sensible depth and tells you what it picked.
Good first questions, in rough order of how much they show you:
- "What does this repository actually do?" — one pass, fast, no changes.
- "Is there a bug in the way we handle X?" — the reviewers will disagree with each other, which is the point.
- "Fix the failing test in
billing." — the lead changes files, the reviewers re-run the suite.
Watch
You will see, in order: a round header naming who leads; the lead working, with its commands appearing as it runs them; then each reviewer, with the commands it ran and a verdict chip saying whether it agrees.
You can stop it at any point. You can also type while it works — your message is picked up at the next turn boundary rather than thrown away.
Read the result
At the end you get one agreed answer and a Summary: the outcome, how many reviewers agreed, how many commands ran and passed, what files changed, and anything the room found along the way but did not fix.
Read the Summary before the answer. It tells you how much to trust what you are about to read.
Then
Rate the answer with the thumbs — it takes a second and it is how we learn which models actually lead well. Ask a follow-up in the same discussion to keep its context. Or press Keep improving if the room ran out of rounds before it settled.