Skip to main content
Tell the API which speaker in a prepared run is your agent, and start the audit. The call returns as soon as the run is accepted — the audit itself takes minutes and runs in the background. Collect the result with audit-run. This is the call that makes a run real. Before it, the run is parked and unbilled; after it, the audit executes whether or not you stay connected.

Authorization

string
required
Your bearer token: Bearer <token>. See Authentication.

Request body

string
required
The run to launch, from audit_prepare.
string
required
Which speaker is your agent. Must be one of the participants returned by audit_prepare, spelled identically. 1–255 characters.

Request

Response

string
The run that was launched.
string
The agent the audit will review. On a repeat call this is the name the first launch set, which may differ from the one you just sent.
string
queued while the audit is still running, completed once it has finished. A first launch is always queued.

Retrying is safe

Launching the same run twice is a no-op success, not an error. If you lose the first response to a timeout, call again: you get 200, the audit is not restarted, not billed twice, and the agent is not changed underneath a review already in progress. The status field tells you where the run actually got to, so a retry after a long silence is also a way to check whether it finished.

Errors

A name that does not appear in the transcript is refused rather than guessed at, so you are never billed for a review of a speaker who does not exist:
Next: Get a run.