Nova + Grok
Grok Voice Agent API Integration
Nova SDK supports xAI's Grok Voice Agent API — the third provider behind the Nova adapter layer. Same SDK, same event surface, selected with a single config value.
Grok Voice configuration
// Configure Nova to use the Grok Voice Agent API
const nova = new Nova({
apiKey: process.env.NOVA_API_KEY,
provider: "grok",
})Nova normalizes Grok Voice Agent events to the same session lifecycle, audio streaming, and event handlers as the OpenAI Realtime API and the Gemini Live API. Provider capabilities differ — see what Grok supports today below.
What Grok supports today
Grok is the newest provider behind the adapter layer. These are the capabilities available in the Nova runtime right now, and where each of the others stands.
Available
- Real-time bidirectional audio
- Barge-in interruption
- Server-side VAD
- Text input
Not on Grok
- Tool and function callingOn the roadmap
- Session resumeOn the roadmap
- File search and RAGAgents needing RAG run on the OpenAI provider
Supported Models
grok-voice-think-fast-2.0Default voice
eve — 28-voice roster
Audio output
24 kHz
Live captions
grok-transcribe
Frequently Asked Questions
Does Nova support the Grok Voice Agent API?
Yes. Nova SDK supports xAI's Grok Voice Agent API as an underlying voice AI provider. Sessions can use Grok Voice instead of or alongside the OpenAI Realtime API and the Gemini Live API — selected at session start with a single configuration value.
Does Centerbeam support xAI?
Yes. Centerbeam's voice agent platform supports xAI's Grok Voice Agent API through the Nova SDK. Grok Voice sessions run on the same Nova event surface as the OpenAI Realtime API and the Gemini Live API, with the provider-specific capability differences listed on this page.
What can Grok Voice do in Nova today?
Grok Voice supports real-time bidirectional audio, barge-in interruption, server-side VAD, and text input. Tool and function calling and session resume are not available on Grok yet. File search and RAG are not available on Grok — agents needing RAG run on the OpenAI provider.
Can I switch to Grok Voice without rewriting my app?
Provider selection is a configuration value, not a code rewrite — Nova normalizes Grok Voice to the same event surface as the other providers. Applications that depend on tool calling, file search, or RAG should stay on the OpenAI provider.