Documentation

Agent Runs

View as Markdown

An Agent run is one execution of an Agent. A run applies the Agent definition to selected variables, source context, model settings, and connected tools, then records the output for review.

Runs can be created manually, triggered by a schedule, started from a webhook, started by an upload trigger, or invoked by another Agent.

Run List

The run list helps you monitor recent and historical execution.

For each run, you can review:

  • Name - the descriptive title of the run
  • Status - new, enqueued, running, completed, failed, or canceled
  • Last Updated - when the run changed most recently
  • Last Completed - when the run last completed successfully
  • Steps - the number of Agent steps
  • Sites - Sites associated with the run

Common list actions include search, filtering by Agent, selecting multiple runs, archiving runs, deleting runs when policy permits it, duplicating a run, and opening run details.

Run Details

The run details page shows the run configuration, progress, outputs, and governance context.

The header includes:

  • run name
  • status badge and running status text
  • start or rerun action
  • cancel action for running runs

The options section shows:

  • parent Agent
  • current status
  • model and reasoning settings, including step-level overrides
  • step count
  • estimated AI Credits
  • notification setting
  • selected Sites and source context

Variables

Variables show the values used for that specific run.

Runs can inherit default Agent variables and override them at execution time. This lets you reuse one Agent for recurring work while changing customer names, date ranges, document sets, report periods, or other run-specific inputs.

Output and Step Results

The output section shows the final result of the run.

Step outputs show the intermediate result from each step, including:

  • step number and name
  • full output content
  • generated files from that step
  • model and reasoning effort used by the step
  • processing timestamps
  • rerun-from-step actions when supported

Use step output review when debugging prompts, validating structured outputs, or proving how an Agent produced a final result.

Generated Files

Agent runs can produce governed spreadsheets, documents, presentations, and images when the selected model and Agent steps support file generation.

Generated files appear with the step output and final output where applicable. They can be downloaded from the run details page and included in governed evidence exports.

Benchmarks

If benchmarking is configured, the run details page can show benchmark output and quality dashboards. Benchmarks can run automatically on completion or manually from the run.

See Benchmarks.

Managing Runs

You can rerun a completed, failed, or canceled run from the run details page. You can also duplicate a run to reuse the same configuration with adjusted variables.

If you need to stop execution in progress, open the run and cancel it. The run status changes to Canceled and the existing record remains available for review.

Governance and Evidence

Agent runs are governed AI records. Records can preserve policy version references, source usage, generated files, step outputs, and final output. Organization administrators export governed evidence from Audit & Governance > Governance Activity.

Evidence bundles can include generated file binaries, metadata, hashes, and signed receipt metadata when signing is enabled.

See Governed AI Records.