Voice Agents

Voice gives an agent the ability to listen and speak. Users can talk naturally and hear spoken responses in real time, enabling hands-free and more accessible experiences. You enable it from the Voice tab of the agent builder.

What voice adds

• Speech in — the user's microphone audio is streamed to the agent.
• Speech out — the agent's reply is synthesised and played back.
• Real-time — audio streams over a websocket so the exchange feels conversational, not request/response.

A voice-enabled agent keeps all its other capabilities — knowledge, tools and workflows — so it can answer from your documents and take actions while speaking.

Enabling voice

1. Open the agent builder and select the Voice tab.
2. Enable voice and choose the voice settings for the agent.
3. Save and publish. Published voice agents expose a real-time voice endpoint that surfaces in the agent's published page and embed options.

Note: Voice uses an audio-capable model. Audio tokens are metered and appear in the usage and cost reports alongside text.

Next steps

• Artifact Generation — produce files, not just speech and text.
• Multi-Agent — coordinate specialist agents.