Text-to-speech copy now describes repeatable narration

text-to-speech has a repeatable narration job: an agent changes a script, keeps the selected voice, and stores the new hosted audio URL with that revision.

Send the text with voice, model, speed, and format. The response returns audio_url plus the settings used, so a later run can reuse them. Text can contain up to 4,000 characters. The endpoint supports mp3, wav, opus, aac, and flac output.

{
  "text": "The release notes now include the migration step.",
  "voice": "af_sky",
  "model": "tts-kokoro",
  "speed": 1,
  "format": "mp3"
}