Run an agent
The Run tab — enter a request, watch the step timeline, approve or answer when the agent asks, and stop.
The Run tab on the agent page. Needs agent.run (Member and up). Limit: 20 runs per minute per
person.
Start a run
- Type the request into Input — attach images if the agent has an image tool.
- Click Run. The button becomes Stop while it runs.
Timeline
Results stream in live:
- Round x of y — the current round against Maximum rounds.
- Every tool call appears as a row going running → done, with its input and output when expanded.
- Warning — for example a tool being unavailable, or the credit ceiling approaching.
- Finally the Output with sources (
[[n]]citations as in Chat) and, if the agent produced files, a Files list to download.
When the agent pauses to ask
Status turns awaiting_input in two cases:
| Situation | You see | You do |
|---|---|---|
| The agent is about to call a changes things tool and Ask before it changes anything is on | An Approval needed block describing the action (email to whom, what gets written) | Approve and continue or Decline |
| The flow has an Ask a person step | A form with the fields that step asks for | Fill it in and Continue |
The run waits indefinitely; you can leave the page and come back through the History tab.
End states
succeeded · failed (open the detail to see the failing step) · stopped (you clicked Stop or
the credit ceiling was hit). Every run is recorded in History with the version it used and the
credits it spent.
Configure an agent
The Configuration tab — instructions, model, knowledge bases, tools, approvals, credit ceiling and memory. Saved with Publish new version.
Flow — multi-step processes
Draw the agent as a graph of steps on a canvas, draft it from a sentence, validate before publishing, and handle a failing step.