Your voice agents.
Your application.

Request a realtime session with your instructions, language and voice. Wait for the model to be ready, then connect using temporary credentials. No saved agent is required.

{
  "instructions": "Help the caller arrange an appointment.",
  "language": "en",
  "voice": "female",
  "reflex_answers": true,
  "steps": null,
  "max_duration_seconds": 600
}

Choose whether to enable reflex answers. Supply a steps graph for a guided conversation, or leave steps empty. These options work independently. Define your own tools, execute requests on your backend, and return results for the conversation to continue. See the documentation for the tool-call flow.

Standalone STT and TTS APIs are coming soon.

We’re updating the infrastructure and architecture before opening standalone streaming speech recognition and generation.

Realtime usage is measured by connected conversation time.

Realtime conversations: ₹2 per minute, billed by actual session duration. Standalone STT and TTS: coming soon.

Existing Swarit Cloud plans remain unchanged.

Our voice-to-voice model for eleven Indian languages and English. Find weights, license information, and release details on the model page.