Skip to main content
An automation is the thing you can run through the API or SDK. It can be implemented by a workflow or by a Git-backed agent, but callers use one run surface:
Use automation targets when you start runs:
Workflow IDs already identify their type, so wf_... needs no workflows. prefix. Slugs still need a prefix because workflow and agent slugs can overlap. For example:
Send this body to POST /api/v1/runs. To select a version, use the version query parameter, for example POST /api/v1/runs?version=1.2.3.

Why the API says automation

The dashboard has workflows and agents because they are authored differently. The public API groups them as automations because they share the same runtime shape:
  • both can be listed with GET /api/v1/automations
  • both can be run with POST /api/v1/runs
  • both produce runs, files, traces, reviews, and evaluator scores
  • both can have datasets and experiments
Use workflow or agent docs when you are authoring behavior. Use automation docs when you are integrating with that behavior from code. Once an automation is running in production, use Reviews to record human feedback and Review production runs to close the loop from a bad run to a verified fix.