Two ways to hear an agent before callers do — and in the Builder the results flow straight back into the chat, so the assistant can fix what was heard.
Test in the browser
In the Builder action bar, press Test (Test draft while a draft is open). A drawer opens with a temporary browser call session.
Try the calls that actually come in: the easy booking, the awkward question, the caller who won't get to the point.
Press End test. The throwaway session is removed; the call is recorded for listening back later.
A browser test can also start from an agent's drawer — Test in browser on its Overview tab. Text-only agents don't take calls, so they have no browser-test button.
Test over a real phone
The telephone test plays exactly what a caller will hear — real network, real handset. In the agent drawer's Deploy tab, choose one of the organisation's numbers or SIP registrations under Telephone test and press Start test — the agent temporarily answers there, and the live transcript streams into the drawer during the call. Closing the drawer tears the temporary deployment down again; nothing is left answering.
Results feed back into the Builder
When a browser test ends in the Builder, the call's transcript and logs go straight back to the assistant, which reviews what happened and proposes fixes to the draft. Confirm, patch, test again — the same loop as troubleshooting a real call.
Drafts answer tests — callers stay live
While a draft exists, every test button talks to the draft, and the test drawer says so: callers stay on the live version. A half-finished redesign can be rehearsed on a real phone line while the published team keeps answering; Go live follows when the tests sound right — see Drafts, versions & going live.
Test calls are real calls — same models, same voices, normal usage. A test that behaved differently from a live call wouldn't be much of a test.