Skip to main content
A voice is a TTS (text-to-speech) voice preset that an agent uses to speak. The catalog is a curated snapshot of voices offered by supported providers — there is no API to create, modify, or delete voices.

Providers and models

GET /v1/voice-providers returns every supported TTS provider together with the models it offers and the tunable settings (sliders) that apply to all voices from that provider. Use this endpoint to populate the provider/model selector in the configuration UI and to render per-provider setting knobs. Each provider entry includes:

Voices and models

For most providers a voice works with any of the provider’s models, so you can change model without changing voice_id. Providers whose voices each belong to a single model list model in filter_facets (currently Rime): there, filter the catalog by model and keep the voice’s own model when you configure it.

Voice catalog

GET /v1/voices returns a paginated list of voices for a single provider. The provider query parameter is required. Each voice entry carries:

Filtering

Pass any combination of gender, accent, age, category, language, or model as query parameters to narrow the list. Each filter performs a case-insensitive exact match against the corresponding field. Voices that lack the field are excluded when that filter is active. The search parameter matches against name or description (case-insensitive substring match). It can be combined with facet filters.

Cursor pagination

The response includes a next_cursor field. When it is non-null, pass it as the cursor query parameter on the next request to fetch the following page. When next_cursor is null, there are no more results. The cursor is an opaque base64-encoded offset — treat it as an opaque string; do not parse or construct it.
limit defaults to 30 and is capped at 100.

Favorites

Users can star voices to keep a personal shortlist across providers. Favorites belong to the signed-in user rather than the workspace, so these endpoints accept session (cookie) authentication only; API keys receive 403.
  • GET /v1/voice-favorites lists the user’s favorites, newest first. Each entry holds the voice as it was listed when starred, plus created_at.
  • POST /v1/voice-favorites stars a voice. The body is the voice entry as returned by GET /v1/voices (without avatar_url). Starring an already-starred voice refreshes its stored copy.
  • DELETE /v1/voice-favorites/{provider}/{voice_id} removes a favorite. Removing one that isn’t starred also succeeds.
A user can keep up to 500 favorites.