Assistant CoreAssistant Core
Voice

Enable Voice

Add voice capabilities to your AI assistant.

Overview

Voice lets users talk to your assistant instead of typing. The assistant listens, understands speech, and responds with natural voice.

This page is for assistant admins. For end-user voice usage, see Web App > Voice Chat. To connect an ESP32 device, see Devices.

Enable Voice

  1. Go to Settings in the sidebar of the Admin Panel
  2. Turn Voice Interaction / Enable Voice on if your workspace or admin role exposes that toggle
  3. When the Voice Settings card appears, configure assistant-level policy:
    • Save Audio Recordings — Store STT/TTS audio for review or debugging
    • Stream Response Text — Send text events alongside audio in voice sessions
  4. Click Save

Models, voices, and the Standard/Realtime mode choice are user preferences in the Web App, not assistant policy in the Admin Panel.

Stream Response Text

The Stream Response Text setting controls whether a voice session sends text alongside audio.

StateWhat users seeBest for
OnUser transcripts, AI response text, the sentence being spoken, and tool-call status textWeb App, demos, support workflows, or cases where users need to read the answer
OffAudio still plays, but voice-session text events are hiddenESP32 or screenless devices, speaker-only kiosks, or low-bandwidth deployments

This setting is available on every plan. Turning Stream Response Text off does not disable voice, does not disable audio, and voice usage still consumes credits normally.

Recommended default: keep it on for the Web App. Turn it off only when the assistant mainly runs on a screenless device or when you do not want transcripts and response text displayed.

Choose Models and Voices for Voice Agent

Users can choose personal model and voice preferences in Web App > avatar > Settings > Voice. Standard mode and Realtime mode have separate model and voice settings.

See Model and Voice Settings to learn how to choose VAD, ASR, LLM, TTS, TTS voice, Realtime model, and Realtime voice.

Voice Modes

Voice supports Standard and Realtime in the Web App. See Standard vs Realtime to compare flow, latency, tool support, model support, and plan availability.

Try Voice

  1. Open the Web App
  2. Click the microphone icon 🎤
  3. Speak your question
  4. The assistant will respond with voice and text

Voice usage consumes credits based on the model and usage. See Credits & Usage for details.

On this page