Skip to content

Run a pipeline manually

This guide shows you how to run a pipeline manually.

  1. Navigate to Pipelines
  2. Find the template you want to run
  3. Click Run (or open the template and click Run pipeline)
  4. If the template has inputs, a form appears — fill them in
  5. Click Start run

The pipeline engine creates a PipelineRun record and begins executing steps.

The run detail view shows:

  • Overall status: running, completed, failed
  • Step progress: each step’s status, input, and output
  • Timeline: when each step started and finished
  • Cost: cumulative token cost across all agent steps
StatusMeaning
PendingWaiting for upstream steps
RunningCurrently executing
CompletedFinished successfully
FailedEnded with an error
SkippedSkipped by a router or skip-propagation

Agent steps accumulate token costs. The run’s totalCostCents and totalTokens are updated as steps complete.

Click any step to see:

  • Input JSON: what was passed to the step
  • Output JSON: what the step produced
  • Error message: if the step failed
  • Duration: wall-clock time
  • Attempt count: how many times the step was tried (if retry policy is set)

If a step fails:

  1. Check the error message in the step detail
  2. If the step has a retry policy, it may retry automatically
  3. If continue-on-fail is set, downstream steps proceed despite the failure
  4. If neither, the run fails

Failed runs can be inspected but not resumed from the UI. Fix the template and start a new run.

Enable Dry run when starting a run to see what would happen without actually executing steps. Dry runs:

  • Evaluate the DAG structure
  • Render template placeholders
  • Show which steps would run and with what inputs
  • Don’t dispatch agents or run scripts