LiveKit integration (voice / video)

FluxyChat voice features use LiveKit as the WebRTC SFU. The Cloudflare Worker issues short-lived JWT access tokens; media runs on your LiveKit server, not on Workers.

Architecture

Browser (livekit-client)  ←WebRTC→  LiveKit SFU
        ↑ JWT token
FluxyChat Worker  (POST /voice/token or dashboard voice-ai routes)

• Worker: token minting with jose (video-voice.js, voice-realtime.ts stubs).
• Dashboard: /voice-ai for agent voice configuration.
• Client: add livekit-client in React apps; connect with token from Worker.

Self-hosted LiveKit

1. Deploy LiveKit server (Docker/K8s).
2. Set Worker secrets:
   • LIVEKIT_API_KEY
   • LIVEKIT_API_SECRET
   • LIVEKIT_URL (wss://your-livekit.example.com)
3. Token mint: POST /admin/calls/token uses mintLiveKitAccessToken in apps/worker/src/lib/livekit-token.js.
4. Extend client apps with livekit-client when connecting to the SFU.

React quick start

import { Room, RoomEvent } from "livekit-client";
import { useLiveKitToken } from "@fluxy-chat/react";

function VoiceRoom({ roomId, adminJwt }: { roomId: string; adminJwt: string }) {
  const { token, fetchToken } = useLiveKitToken({
    workerUrl: process.env.NEXT_PUBLIC_FLUXYCHAT_WORKER_URL!,
    adminJwt,
    roomId,
    displayName: "Alice",
  });

  async function join() {
    const creds = await fetchToken();
    if (!creds?.token || !creds.url) return;
    const room = new Room();
    await room.connect(creds.url, creds.token);
  }

  return <button type="button" onClick={() => void join()}>Join voice</button>;
}

LiveKit Agents (optional)

For AI voice agents, use LiveKit Agents as a separate process. FluxyChat agent runtime can hand off to an agent worker via webhook when a voice session starts.

References

• LiveKit overview
• Access tokens
• Worker stubs: apps/worker/src/lib/video-voice.js, apps/dashboard/app/voice-ai/