VoiceVoice agent
Voice agent
Talk to a live agent over WebRTC: speech recognition, a language model, then a voice.
Ready
Sign in to try the playground · Usage spends your workspace's credits. Developers can paste an API key instead.
Ready
Choose a model and voice in Parameters, then start a session and speak.
Voice credit meter
Sessions bill STT audio, LLM tokens, and TTS speech. Keep test calls short.
Pick models in Parameters, then start a live session.