speech-to-speech Built by Hugging Face

Open sourceLive demo

Voice agents you can pick up and call

Choose a scenario, press call, and just talk. Interrupt whenever you like.

The open models behind every call

  1. Speech detection Silero VAD
  2. Turn taking pipecat-ai/smart-turn-v3
  3. Transcription nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-NVFP4
  4. Language model nvidia/Gemma-4-31B-IT-NVFP4
  5. Voice Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice

Chained together by speech-to-speech and served on Hugging Face Inference Endpoints. Your microphone is only used while a call is active.