LiveKit Agents run Grok Voice for speech-to-text, text-to-speech, and speech-to-speech on existing real-time audio infrastructure.
Grok Voice models are now available on LiveKit Agents. Speech-to-text, text-to-speech, and speech-to-speech models can now run within the same LiveKit infrastructure teams already use for real-time audio.
Build voice agents that deliver real customer outcomes using Grok and deploy them on LiveKit infrastructure.
LiveKit and SpaceXAI have built a joint patient intake agent to represent the front desk of medical clinics: xai.livekit.space.
The agent can book a visit, handle rescheduling and cancellations, and answer a policy question in the middle of another task without losing sight of the end resolution. Under the hood, it fires tool calls without slowing the call or altering the naturalness of the conversation.
The entire agent is published as a LiveKit recipe. LiveKit Agents can run Grok speech-to-speech natively; this recipe uses a cascaded stack (speech-to-text → language model → text-to-speech) all through LiveKit. Swap the fake clinic for a real EHR or scheduler. The rest of the agent doesn't change, including over the phone via SIP.
Grok Voice models run through LiveKit Inference, so an existing LiveKit agent can use them directly within LiveKit.
See current API pricing. The LiveKit docs cover the integration, or start from the patient-intake recipe and swap in your own backend.
