Close a conversation

Open Blueprint → Conversation completion. Choose what happens after the assistant has replied and the visitor stops responding.

Choose an inactivity policy

Do nothing leaves the conversation open.

Cancel immediately closes it when the wait expires, without a completion model call or output.

Try to complete, otherwise cancel uses a model call after the wait to judge whether the task finished and extract your output fields. If the task is unfinished, required data is missing or confidence is too low, the conversation closes without output.

Set Wait duration and its Time unit. New activity resets the timer.

Describe a successful outcome

Write Completion criteria that the conversation can prove. For example:

The visitor supplied an order code and said goodbye.

Set a Confidence threshold between zero and one. Choose the Completion model that runs this check.

Under Output fields, choose Add output field. Enter its name, data type and extraction instructions. Keep Required checked if completion needs that value, then choose Save field.

For example, add a text field named order_code and describe it as “The order code explicitly supplied by the visitor.” Successful completion returns that code in the output. With no fields, successful completion returns an empty object.

Use Advanced JSON for an output schema that needs more detail than the field form.

Test both outcomes

After the draft saves, reset the Test Console. Supply the required information, say goodbye, and wait for the policy to run. Expand Completion judgment to inspect the decision, confidence, reason and output.

Reset again and leave out a required value. Check that the agent doesn’t invent it to complete the task.