You can try an agent two ways from the flow editor: have a real conversation with it in a preview, or run the draft with a sample payload and read the transcript. Neither reaches a real customer.
Preview a conversation
Preview opens a web chat test panel next to the canvas.
- 1Open in a new tab
Select Preview in the editor's top bar. If you have the
builder.bots:triggerpermission and the agent has unpublished changes or has never been published, the button reads Preview draft and the chat runs your draft, marked with a Draft preview badge.Chat as a customer would. The panel's note says this is a test session: nothing you say reaches your inbox, your contacts or your reports.
To work with more room, select Open in a new tab.
Without the trigger permission, Preview runs the published flow. It is disabled, with the hint Publish this flow first to preview it, until the agent has been published once. Preview is also on each card in the Agents list.
Run the draft
Test (visible with builder.bots:trigger) opens Test this flow. It runs the current draft without a live conversation. Natiq saves the canvas first, and nothing reaches a real channel or contact.
Select Test.
Optionally fill in Trigger payload, a JSON object. Question steps that the payload covers are pre-filled. Select Autofill from flow to add a blank key for every question's Saves to variable.
Decide on Skip integration steps. Turn it on to run app steps as no-ops so nothing is sent to a third party. Natiq remembers your choice.
Select Run the draft.
The result shows a badge and a transcript.
| Badge | Meaning |
|---|---|
| SUCCEEDED | The run finished without a step failing. |
| STEP FAILED | The run finished, but a step reported a problem. The failing step is named. |
| HALTED | A step was set to stop the flow on failure, so the steps after it never ran. |
| FAILED | The run could not finish. |
The transcript lists what the bot said and what the engine did: variables pre-filled from the payload, app steps that ran or were skipped, steps you turned off, and where a Handover would have happened. Raw result shows the full output.
If a question has no answer in the payload, the run fails and Natiq lists the variables it still needs.
A headless run is not a chat
Because there is no live customer, a Handover step is only noted in the transcript, and a flow with no customer-message trigger cannot use one at all. Use Preview to feel the conversation and Test to check the logic.
Where test runs show up
Runs started by Test are listed with the trigger Test in Analyze → Runs. See Agent analytics.