Using the coding agent in Code

Start a project

Open Code and create a project from a template (blank, Python, Node.js or website), or import one of your GitHub repositories.

Your files are saved to your account. The project's computer — where commands and the preview run — starts only when you, or the agent, need it.

Describe the change

In the Agent panel, say what you want, as you would to a colleague: “Add tests for the cart and make them pass”, or “Fix the crash when the list is empty”.

Choose whether to see the plan first (on by default), whether to review each set of edits before it's saved, and which commands may run without asking.

Approve, answer, watch

The agent reads the relevant files and proposes a plan. Nothing changes until you approve it.

It then edits files and runs tests or builds. Commands outside the safe list (tests, builds, installs and read-only commands) wait for your OK; deleting everything, deploying and git push never run.

If it needs the computer, it asks you to start it. Starting it uses workspace minutes from your plan.

Review and revert

When it finishes you get a summary of what it changed, what it ran and what's left. Open Changes to see every edited file as a diff.

Revert a whole run or a single file. If you changed a file since, Orbis warns you before overwriting it.

What it costs

Each step the agent takes counts like one chat message on your plan; the run shows its step count. Commands use workspace minutes, counted in 10-minute blocks.