Agents & Workflows
Agents work on an initiative, bug, or todo for you, and report progress and results back to that item. Workflows chain agent steps and human review into a repeatable process.
Where to start
| Goal | Start here |
|---|---|
| Run your first agent with Desktop | Quickstart |
| Build and publish a repeatable process | Workflows |
| Configure the agents running on your machine | Local agents |
| Run agents without Desktop | Use in terminal |
| Inspect queues, runs, reviews, and agent health | Monitoring |
| Build your own coding agent | Building Agents |
Queued agents or a coding tool?
Send to… either opens a coding tool on your machine with a prefilled prompt, or queues the work for an agent or workflow. When you open a coding tool, you drive it. When you queue work, an agent runs it and reports progress back.
How queued work runs
Queuing work creates an agent session. An available agent claims the session, starts work, reports activity as it goes, and finishes with a result: completed, failed, or released. A workflow run can create several sessions as work moves between agent steps and review.
Everything stays on the initiative, bug, or todo: the spec, comments, documents, status, and result. The agent treats that content as context, not as permission. Its sandbox, credentials, and access limits come from the agent's configuration.