Agents

Resume an agent run waiting on human input

When an agent reaches a human-input step, it pauses and waits for a response. You can resume the run by replying directly in chat. The agent stores your response, marks the step as succeeded, and continues to the next step.

When an agent pauses for input

An agent pauses when a step is configured to collect human input before continuing. The run detail page shows a Waiting for input panel, and the run status is waiting. While a run is waiting, it is distinct from a normal chat conversation. The chat reply is treated as agent input, not as a new assistant query.

Resume a waiting run from chat

  1. Open the agent run detail and confirm the status is waiting and the Waiting for input panel is visible.

  2. Click Reply in chat to open the chat panel.

  3. Type your response exactly as the waiting prompt requests. The reply must match the expected input for the step.

  4. Send the message. The agent resumes automatically if the reply matches the waiting prompt.

  5. The step is marked succeeded, the response is stored in the agent state, and the run proceeds to the next step.

What happens after you reply

Once the agent resumes, it continues executing the remaining steps. If another human-input step appears later, the run pauses again and repeats the same process. When all steps complete and the output contract is satisfied, the run finishes with the configured deliverable.

Why a resume might fail

A resume attempt fails if any of these conditions are true:

  • The run is no longer in a waiting state.

  • The run does not have a waiting step with an exposed input key.

  • The waiting step no longer expects input.

  • The response is empty.

If the run is not waiting, check the Run History to see whether the run already completed, failed, or was resumed by another user.

Was this helpful?
Was this helpful?