Production checklist

​

Security

• The API key is in your server’s secrets (for example TALKING_AVATAR_API_KEY), never in browser code, a mobile app or a repository.
• Your /avatar route is behind your app’s authentication, so only your users can watch a conversation.
• You use a separate API key per environment (development, staging, production), so you can revoke one without affecting the others.

​

Lifecycle and billing

• There’s exactly one LiveAvatar per conversation.
• You call close() when the conversation ends, the user leaves, or anything fails.
• You handle LiveAvatarError from start() and the close event, and your voice agent keeps working without the avatar.
• Auto top-up is on, or you watch your balance, so sessions don’t stop with 4002.
• A monthly spending limit caps what a bug could cost.

​

Audio

• The sampleRate matches your TTS audio exactly.
• You send raw PCM16 mono, not MP3, Opus or WAV with headers.
• The voice plays only through the avatar, never twice.
• Barge-in calls clear(), and your agent’s transcript is truncated to heardMs.

​

The page

• connect() runs from a click or tap, so the browser lets the voice play.
• You handle the muted status: show a button that calls view.unmute().
• You disconnect() the view when the conversation UI goes away (for example, on React unmount).
• The container has a size. The 16:9 video fits inside it without cropping.

​

Networks

• For users behind strict corporate firewalls, you pass your own TURN server in iceServers:

new LiveAvatarView({
  container,
  endpoint: "/avatar",
  iceServers: [
    { urls: "stun:stun.example.com:3478" },
    { urls: "turn:turn.example.com:3478", username: "user", credential: "secret" },
  ],
});

​

Disclosure

• Where the law requires it, and whenever people might otherwise be misled, you tell your users that they are talking to an AI. See the Terms of Service.