Nova Realtime
Real-Time Voice Agent Runtime
Nova Realtime is the runtime layer that handles bidirectional audio, turn-taking, VAD, and provider routing for every Nova voice agent session.
Runtime capabilities
Sub-500ms turn-taking
Designed for the latency requirements of operational work — not conversational demos.
Industrial VAD
Voice Activity Detection tuned for noisy environments: factory floors, cleanrooms, outdoor field sites.
Barge-in interruption
Users can interrupt the agent mid-response. The runtime cancels cleanly without audio artifacts.
Bidirectional audio
Full-duplex WebSocket audio stream. Listen and speak simultaneously.
Multi-provider routing
Switch between OpenAI Realtime and Gemini Live per session without application changes.
Edge deployment
The relay component can run on-prem or at the edge with no external network dependency.