Configuration overview

Voice Agents use a dedicated configuration screen, separate from the standard agent configuration UI. The screen is organized top-to-bottom in the order you should fill it out:

1. Identity — name, description.
2. Environment Variables — keys, values, masked secrets. Same model as standard agents.
3. Mode Selection — the execution path. See Execution modes.
4. Model Configuration:
   • LLM — provider and model for the reasoning step.
   • Transcriber (STT) — provider, model, language, and advanced options. See Transcriber (STT).
   • Text-to-Speech (TTS) — provider, voice, voice settings, and fallback voices. See Voice (TTS).
5. Voice Settings — Voice Activity Detection, interruption, tool announcements. See Voice Activity Detection.
6. Base instruction — the system prompt for the voice agent.
7. Knowledge Base — attach one or more KBs. See Knowledge Bases.
8. Tool Selection — voice-eligible tools. See Tools.
9. Additional Features — voice-compatible features only.

What's not on the screen

These options exist for standard agents but don't appear on the voice agent configuration:

• Guardrails — not yet configurable on voice agents.
• Planning and sub agents — not yet supported on voice agents.
• Browser Use, Artifacts, Reflection, Follow-up Messages — not applicable to a voice interaction; hidden under Additional Features.

BYOK is everywhere

LLM, STT, and TTS providers all support BYOK (bring your own key) through a unified key selector. See BYOK for the key management UX and cost implications.

Multilingual

If you select a multilingual ASR (Deepgram Multilingual or Google Multilingual), the TTS provider list filters automatically to providers that support multilingual synthesis. See Multilingual.

Saving and publishing

The voice agent uses the same draft / publish / version-diff model as the rest of the platform. See Versioning for the shared mechanics — version history and configuration diff between versions work identically on voice agents.