Generate Starting Message Audio
Pre-generate static-greeting audio for a bot_initiates_static opening.
Synthesizes request.text with the version’s configured TTS voice and
stores the MP3 in the recording S3 bucket under a content-addressed key
(org + a fingerprint over the text + full TTS config). Returns a
short-lived presigned URL for immediate builder playback. The builder
then sets starting_message.pregenerated_greeting_audio = true; at
runtime the agent recomputes the same fingerprint from the rendered text
- TTS config and plays the stored file instead of synthesizing TTS.
Pre-generation is only supported for non-templated text: the runtime fingerprint is over the rendered greeting, so a Jinja greeting would never match a stored object.
text comes from the request so it works on unsaved drafts; the voice
is read from the saved version’s primary TTS model.
Headers
Organization ID for user PATs (pat_ prefix tokens)
Body
Request to pre-generate (or re-preview) static-greeting audio.
text is supplied by the caller so it works on unsaved drafts (mirrors
suggest-action-grid). The TTS voice is read from the saved version's
configured TTS model, and the stored object is content-addressed over
text + that config, so the same body identifies the same recording.
Response
Successful Response
Short-lived presigned URLs for playing/previewing greeting audio.
An agent with a randomized voice pool has one recording per pool voice — any of them can open a real call, so all of them are auditionable.
The primary voice's recording (pool entry 0 when there is a pool).
Serialized alongside recordings so a client that predates the
per-voice list keeps working — including across a deploy where the
backend rolls out ahead of the builder.

