Workflows

Execution model

1. A task describes the work.
2. The dispatcher promotes it into a run, with scheduled retries and a watchdog for stalled runs.
3. The agent executes using its configured model, tools, and skills — via native agent sessions.
4. High-risk capability calls pause for human approval (architecture; enforcement depth expanding).
5. The result is recorded as an outcome with a status you can inspect.

Multi-step pipelines (chaining runs into repeatable workflows) are available as an experimental capability in the app.