First run

Start Studio

Start the embedded local web application. It can manage sessions in any repository, regardless of the directory where you launch it.

Studio prints its loopback URL and opens it in your browser. Keep that q process running while you use the interface.

Configure a model

Open Settings → Providers, add a Gateway provider, and enter its endpoint and API key environment variable. Then open Settings → Models and assign a discovered model to Default. Provider and model changes save automatically.

Use Settings → System One only when you want a separate typed-decision provider for tasks such as Agent Skill relevance checks.

Create a session

Open Sessions and choose Add session. Select the repository directory with the folder browser, then choose an existing root session or create a new one. Studio remembers the repository location for that root session.

Starting Studio in a repository subdirectory does not widen the file boundary. Each session runs in the exact workspace directory registered for it.

Send the first request

Begin with a concrete repository question so you can see how q gathers evidence:

For a change that benefits from a specialist, ask the main agent to delegate or explicitly name an occupational role:

Child agents appear below their parent in the session tree. Select a child to inspect its own transcript and tool activity.

Review the result

Open Changes to browse staged, unstaged, renamed, and untracked repository changes. Select a file for its highlighted diff.

When the diff is ready, start commit review to generate a Conventional Commit or split-commit proposal. Nothing is committed until you execute the selected proposal.