Introduction to voice interaction

What voice enables

• Talk to the agent: hold the microphone, record a request, and convert it to text.
• Listen to responses: turn a written response into audio using the configured voice and speed.

Voice is available in Studio tests and the Web Widget when enabled.

How a conversation is processed

Audio is transcribed, then sent to the same agent, knowledge, and tools as written chat. The response is generated as text and synthesized when the user listens. Instructions, memory, knowledge, and tools behave identically.

Current scope

Voice complements Studio and Web Widget. It is not a telephone, numbering, VoIP, or SIP channel.

Enable voice only after the agent works reliably in text.