floch-rt demo
Rust media + session runtime · WebRTC · SSE
Provider
OpenAI Realtime (gpt-realtime)
Gemini Live 2.5 Native Audio
xAI Grok Voice
Groq Llama 3.3 70B → Deepgram Aura (fastest LLM)
OpenAI GPT-4o-mini → Deepgram Aura
Gemini 2.5 Flash → Deepgram Aura
Gemini direct audio (skips STT) → Deepgram Aura
Cascading mode trades ~200ms extra latency for lower cost and more model flexibility. STT + TTS are both Deepgram for all variants.
Model
Voice
System prompt
You are a helpful voice assistant. Keep responses concise and conversational.
Greeting
(optional — agent speaks first)
Click to start
--
TTFT (ms)
0
Turns