Voice AI · Conversational Assistant

Conversation

Start a conversation to see the transcript here.

Operator Control

Preset / Persona

Agent Name

TTS Voice

VAD Threshold 0.65 Higher = less sensitive (reduces false triggers)

Use my GPU (browser runs LLM – faster, needs WebGPU) Runs LLM in your browser with your GPU. First load downloads ~150MB (SmolLM2-360M). Needs WebGPU.

LLM Provider Groq: free at console.groq.com. Set VOICE_AGENT_GROQ_API_KEY on server.

LLM Model Ollama: ollama pull llama3.2

End-of-turn pause (ms) 400 Higher = wait longer before replying (reduces mid-sentence cuts)

Temperature 0.55

Greetings

Closings

Developer key

Clone my voice (AI speaks as you)

Guided setup walks you through 6 sentences for best results. Or record freely (6+ sec) / upload a file.

Debug

VAD

Level

State idle

STT — ms

LLM — ms

TTS — ms