@ariaflow/livekit-plugin-transport-twilio

Twilio Media Streams transport adapter for AriaFlow voice agents. Works with Node.js, Bun, Deno, and any WebSocket-enabled runtime. Cloudflare Workers support is currently unstable.

Features

• ⚠️ Cloudflare Workers Compatible (Unstable) - No Node.js dependencies
• ✅ Twilio Media Streams Support - Native integration with Twilio's API
• ✅ Automatic Codec Conversion - G.711 μ-law ↔ PCM with resampling (8kHz ↔ 24kHz)
• ✅ AriaFlow Native - Works with createAriaFlowSession and AriaFlow Runtime
• ✅ Session Management - Full lifecycle management with SessionManager

Installation

# Using npm
npm install @ariaflow/livekit-plugin @ariaflow/livekit-plugin-transport-twilio

# Using bun
bun add @ariaflow/livekit-plugin @ariaflow/livekit-plugin-transport-twilio

Quick Start

Cloudflare Workers (Unstable)

⚠️ Note: The Cloudflare Workers integration is currently unstable and not officially supported. Use at your own risk.

// wrangler.toml:
// name = "ariaflow-twilio-agent"
// compatibility_date = "2024-01-01"
// compatibility_flags = ["nodejs_compat"]

import { createTwilioWorker } from '@ariaflow/livekit-plugin-transport-twilio/cloudflare';
import { createAriaFlowSession } from '@ariaflow/livekit-plugin';
import { GeminiLiveSTT, GeminiLiveTTS } from '@ariaflow/livekit-plugin/gemini';

const agentConfig = {
  agents: [{ id: 'assistant', prompt: 'You are helpful.' }],
  defaultAgentId: 'assistant',
};

export default createTwilioWorker({
  agent: () => createAriaFlowSession({
    runtime: agentConfig,
    stt: new GeminiLiveSTT(),
    tts: new GeminiLiveTTS(),
    greeting: 'Hello! How can I help?',
  }),
});

Node.js / Deno / Any WebSocket

import { WebSocketServer } from 'ws';
import { TwilioTransportAdapter } from '@ariaflow/livekit-plugin-transport-twilio';
import { createAriaFlowSession } from '@ariaflow/livekit-plugin';
import { SessionManager } from '@ariaflow/livekit-plugin';
import { GeminiLiveSTT, GeminiLiveTTS } from '@ariaflow/livekit-plugin/gemini';

const wss = new WebSocketServer({ port: 8080 });
const sessionManager = new SessionManager();

const agentConfig = {
  agents: [{ id: 'assistant', prompt: 'You are helpful.' }],
  defaultAgentId: 'assistant',
};

wss.on('connection', (ws) => {
  const transport = new TwilioTransportAdapter({
    send: (msg) => ws.send(msg),
  });

  ws.on('message', (data) => {
    transport.handleMessage(data.toString());
  });

  ws.on('close', async () => {
    await transport.close();
  });

  // Wait for 'connected' event before starting session
  ws.once('message', async (data) => {
    const event = JSON.parse(data.toString());
    if (event.event === 'connected') {
      const { agent, sessionOptions } = createAriaFlowSession({
        runtime: agentConfig,
        stt: new GeminiLiveSTT(),
        tts: new GeminiLiveTTS(),
      });

      await sessionManager.startSession(transport, agent, sessionOptions);
    }
  });
});

TwiML Setup

Create a TwiML bin in your Twilio console:

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Say>Connecting to AI assistant...</Say>
  <Connect>
    <Stream url="wss://your-worker.workers.dev/voice/stream" />
  </Connect>
</Response>

Audio Specifications

The transport handles all codec conversion and resampling automatically.

API Reference

TwilioTransportAdapter

Main transport adapter class.

const transport = new TwilioTransportAdapter({
  id: 'optional-custom-id',
  send: (message: string) => {
    // Send message to WebSocket
    ws.send(message);
  },
});

Properties:

• id: string - Unique identifier for this transport
• audioInput: TwilioAudioInput - Audio input stream
• audioOutput: TwilioAudioOutput - Audio output stream
• textOutput: TwilioTextOutput - Text output stream
• config: TransportAdapterConfig - Configuration object
• isOpen: boolean - Whether the transport is active

Methods:

• handleMessage(message: string) - Process incoming Twilio message
• clearAudio() - Clear audio buffer
• close() - Close the transport

createTwilioWorker(options) (Unstable)

⚠️ Note: This Cloudflare Workers integration helper is currently unstable and not officially supported.

Cloudflare Workers integration helper.

export default createTwilioWorker({
  agent: () => createAriaFlowSession({...}),
});

Options:

• agent: () => Agent | { agent: Agent; sessionOptions: any } - Agent factory
• stt?: any - STT instance (for raw LiveKit pattern)
• llm?: any - LLM instance
• tts?: any - TTS instance
• vad?: any - VAD instance
• turnDetection?: any - Turn detection config

Environment Compatibility

Limitations

1. Twilio Sample Rate: Twilio uses 8kHz G.711 μ-law. The transport resamples to 24kHz for AriaFlow, but this adds slight latency.
2. Codec: Only G.711 μ-law is supported (Twilio's standard).
3. Latency: Expect 500-1000ms additional latency compared to direct WebRTC.
4. Full Duplex: Twilio Media Streams is full-duplex, but the implementation treats inbound/outbound as separate streams.

Troubleshooting

Audio cutting out

• Check your WebSocket connection is stable
• Verify send callback is working correctly
• Check for errors in the console

Poor audio quality

• The 8kHz sample rate is lower than typical 24kHz
• This is a Twilio limitation, not the transport
• Consider using Twilio's programmable voice SDK for higher quality

Connection issues

• Ensure your WebSocket URL is publicly accessible
• Check Twilio can reach your server (no firewall blocking)
• Verify CORS headers if using browser-based testing

Examples

See examples/ directory for:

• cloudflare_worker.ts - Complete Cloudflare Worker example (⚠️ Unstable)
• wrangler.toml - Deployment configuration

License

Apache-2.0

Related Packages

• @ariaflow/livekit-plugin - Core AriaFlow LiveKit integration
• @ariaflow/livekit-plugin/gemini - Gemini STT/TTS plugins
• @ariaflow/livekit-plugin-transport-ws - WebSocket transport
• @ariaflow/livekit-plugin-transport-http - HTTP/SSE transport