Twilio

Syrinx terminates Twilio Media Streams directly — the μ-law 8 kHz phone audio is transcoded to PCM16 for your pipeline, and your agent’s speech is transcoded back, all inside one WebSocket per call.

Point a number at your agent

Configure your Twilio phone number’s “A call comes in” webhook to POST https://<your-host>/incoming-call. Syrinx returns TwiML that bridges the PSTN leg to the /twilio media-stream endpoint:

<Response>
  <Connect>
    <Stream url="wss://<your-host>/twilio?sessionId=<callSid>" />
  </Connect>
</Response>

The Twilio CallSid becomes the session id, so the call’s media, resume, and any recording all key off the same stable id.

On a Node server

Mount the Twilio media-stream handler alongside your other routes; it speaks the Media Streams protocol (connected / start / media / mark / stop) and drives a VoiceAgentSession per call.

On Cloudflare Workers

The reference Worker binds a TwilioVoiceConversation Durable Object to /twilio — see Deploy on Cloudflare. One hibernatable Durable Object per call, no server to run.

Codecs

Twilio Media Streams is μ-law (PCMU) at 8 kHz in both directions. Syrinx handles the transcode; there’s nothing to configure. For wideband audio (G.722) you’ll want Telnyx.

See Codecs, DTMF & transfer for sending DTMF and transferring calls.