Skip to main content

Bulk Run

Bulk Run processes a batch of inputs through a published application (Tool or Agent) as a single managed job, instead of calling it once per input. Use it for back-testing a prompt against hundreds of rows, enriching a spreadsheet, classifying a backlog, or any "run the same app over many inputs" task.

How it works​

  1. Open a published version of the Tool or Agent. (Bulk Run operates on published applications so the version is stable for the whole batch.)
  2. Choose Bulk Run and provide the batch of inputs — typically a dataset or an uploaded file (CSV/XLSX) where each row is one run.
  3. Map columns to inputs — match each application input to a column in your file.
  4. Start the job. The batch is processed through a dedicated execution queue, separate from interactive single runs so it doesn't block them.
  5. Track progress as rows complete, then download the consolidated results (each row's output alongside its input).

Resilience​

Bulk jobs are designed to survive infrastructure events. If a pod restarts or a deployment rolls out mid-batch, the job resumes from where it left off rather than restarting or dropping rows — so large batches complete reliably.

Credits, guardrails, and tracing​

  • Each row consumes credits exactly as a single run of the same application would; the bulk job's cost is the sum of its rows.
  • Guardrails configured on the application apply to every row.
  • Every row produces its own trace in Run history, grouped under the bulk job, so you can inspect any individual run.