Run coding agents through a graph

Orbital runs a coding agent through steps you write in a DOT graph. Each agent node supplies a prompt. Edges select the next step using the agent’s structured answer or a machine observation, such as a pull request’s state.

Use a graph when the work repeats, needs review, or waits for something outside the agent. You can inspect the run, steer it and resume stopped work in your browser. A short one-off question usually needs no graph.

Start with your first run. Then learn how runs work or copy a complete example.

Orbital runs locally. The server owns execution and history; the browser controls it. This documentation site is static and works without an Orbital server.