voice_agent

Add an AI participant that listens to room audio and responds with synthesized speech.

The audio output track appears when the first synthesized audio packet arrives. It preserves the provider’s sample rate and channel layout, and idle periods are represented as timestamp gaps rather than generated silence. Add an audio_mixer when a continuously clocked output is required.

Example

{
  "name": "bot",
  "type": "voice_agent",
  "inputs": ["mix_audio"],
  "config": {
    "language": "en-US",
    "instructions": "You are a helpful live show co-host.",
    "greeting": "Hello everyone!",
    "interrupt": true
  }
}

Config

Field	Type	Default	Description
language	string	managed default	Optional BCP-47 conversation language hint, such as en-US or zh-CN. When omitted, the managed service uses language detection or its default language behavior.
instructions	string	—	System prompt steering agent behavior. Truncated to 8000 characters.
greeting	string	—	Optional first utterance spoken when the agent joins. Truncated to 2000 characters.
voice	string	managed default	Optional voice id supported by the managed service.
interrupt	bool	true	Barge-in: when a participant starts speaking, the agent stops and pending synthesized output is discarded. Set false to always finish the turn.

Pricing

$0.10/min while the voice-agent session is connected. Node pricing.