Workflows

Testing and run logs

Test a workflow draft in the Run tab, run it over a table of inputs with Batch, and find past runs in History.

Open in ChatGPT Open in Claude

You test a workflow in the editor before you publish it. The Run tab sends a message through your draft, the Batch tab runs the workflow once per row of a table, and History shows what happened in past runs.

Tests are real runs

A test runs the draft’s steps for real. It spends credits and uses connected accounts, so a step that sends an email or posts to Slack does it. Point test runs at safe targets.

Try the draft in the Run tab#

Click Run in the switch at the top of the editor. A dialog opens with the workflow’s name and your earlier test chats on the left, and the chat on the right.

  1. Type a message and send it. The workflow starts from its chat trigger with your message as input.
  2. Read the answer in the chat. Each conversation is saved as a session, and the one you are in is listed as Current session.

The Run tab always uses the draft, so you see your latest edits without publishing. It does not run the live version. If a workflow has a Session Closed trigger, the dialog offers Close Session Now so you can fire it without waiting for the session to expire.

The Run and Batch buttons and Publish are disabled when the workspace has no active subscription or not enough credits. Hover over a disabled button to see why. See Credits and the out of credits page.

Read an earlier test#

Pick a session in the left list to read it again. When you view a past session, a Plain and Verbose switch appears above the conversation. Verbose shows more detail about what happened during the run. If a run fails, see Run failed.

Run many inputs with Batch#

Batch runs the workflow once for each row of a table. It is useful for checking a change against many examples, or for processing a list.

  • Click New Batch, then add rows by hand or choose Import CSV. The first row of the file becomes the column names.
  • Set Concurrency to decide how many rows run at the same time.
  • Use Run Pending to start the rows that have not run, Re-run Failed to retry failures, or Re-run All to start over. Stop All halts the run.
  • Each row shows a status of pending, running, completed or failed. Open a row to read its output and error, or use View session history to see how it ran.
  • Export CSV downloads the results.

Batch is available on the Trial, Pro, Agency and Enterprise plans. On other plans, clicking Batch opens the plans dialog.

Find runs outside the editor#

The Run tab only shows your own test chats. Conversations that visitors have with a chatbot, and runs started by a schedule, are recorded in History in the sidebar. Use the Agent or workflow filter to narrow the list to one workflow, open a session and switch between Plain and Verbose. The Chat logs tab lists individual events and has a search box.

A scheduled run has nobody watching, so History is the place to confirm it worked. See Schedules.

Test from an AI agent#

If you connected an AI assistant to AI Studio, it can run your draft too, with the same cost and effect as the Run tab. See Connect your AI.