Realtime voice
Aven's production WebRTC path. This runs alongside the browser-speech fallback while the realtime agent stack is rolled out.
Voice engine: LiveKit (Deepgram · LLM · Inworld) on the Azure worker. If Aven connects but doesn’t speak, confirm LiveKit Inference/billing is enabled.
Talk to Aven
Ready for a realtime conversation
WebRTC audio via LiveKit. You can interrupt naturally once the agent worker is running.
What this path proves
- Authenticated short-lived session tokens
- Browser microphone capture over WebRTC
- Remote Aven agent audio playback
- Per-user session isolation
- Streaming turns, tools, and barge-in
