Deployment
Avatar video
Live avatar video sessions with Tavus or HeyGen (LiveAvatar), driven by your agent's knowledge, tools, and guardrails.
Avatar video
Give your agent a face. With avatar video enabled, visitors can start a live video session in the web widget and talk face-to-face with a lifelike avatar that answers with your agent's brain — the same knowledge base, tools, and guardrails as chat and voice.
Providers
WisebotAI supports two avatar providers, connected with your own API key (BYOK — media costs are billed by the provider to your account, not marked up):
| Provider | Connect at | What you need |
|---|---|---|
| Tavus | Dashboard → Avatar Video | A Tavus API key and a replica (face) ID |
| HeyGen (LiveAvatar) | Dashboard → HeyGen Avatars | A LiveAvatar API key from app.liveavatar.com and an avatar ID |
Connect the key once per organization; it is encrypted and stored securely. Only admins should manage these values.
Enable video on an agent
- Open Agents → your agent → Modalities.
- Toggle Video avatar on and pick a provider.
- Enter the provider-specific ID — Replica ID for Tavus, Avatar ID for HeyGen — plus optional voice, language, and quality settings.
- Save. The widget's "Start a video session" button appears automatically once the provider key is connected and the agent has video enabled.
How sessions work
- Agent brain: by default the avatar's answers are generated by your agent — knowledge search with citations, tools, and guardrails included — not by the provider's stock model. Turn this off per agent with the Use agent brain toggle (Tavus only; HeyGen sessions always use the agent brain).
- One audio source: in a video session the avatar owns the audio; voice TTS is muted so customers never hear two voices.
- Transcript: the conversation transcript is captured live and saved to the conversation when the session ends, so your team sees video sessions in the same inbox timeline as chat and voice.
- Session limits: sessions are capped (30 minutes) and end automatically after inactivity; customers can pause and resume without losing the transcript.
Testing
- Tavus offers a test mode and LiveAvatar a sandbox mode — both let you exercise sessions without consuming provider credits; check your provider dashboard for how to enable them on your account.
- Avatar rendering takes a few seconds after the session connects; the widget falls back gracefully if video is unavailable.
Related
- Voice assistant for phone- and voice-based experiences
- Widget customization for the embed surface where video sessions run
- Built-in AI tools for what the agent can invoke mid-session