Twilio Agent Connect ships GPT-Live-1 bridge for phone agents
TL;DR
- Twilio's September 10 Python and Node.js tutorials wire OpenAI's GPT-Live-1 directly to Programmable Voice through the Agent Connect SDK's GPTLiveProvider.
- GPT-Live-1 splits work across two models: a speech-to-speech model handles audio, and a separate reasoning model handles tool calls.
- The Node.js reference stack uses Fastify, PCmu audio at 8000 Hz, and the 'Marin' voice, with tool-calling delegated to gpt-5.6-terra.
A YouTube tutorial on wiring OpenAI's GPT-Live-1 to Twilio for a voice agent lands alongside Twilio's own documentation push for the same stack on September 10.
In the Python walkthrough, Twilio's Xinghao Huang describes how Agent Connect "ships a GPTLiveProvider that bridges Twilio Programmable Voice Media Streams directly to GPT-Live-1," letting developers point an inbound phone number at a speech-to-speech model without writing WebSocket plumbing. A parallel Node.js version from editor-in-chief Paul Kamp runs on Fastify, pipes audio at PCmu 8000 Hz with the "Marin" voice, and delegates tool-calling to a separate reasoning model.
Kamp's framing is the clearest on how the work is split: GPT-Live "runs a conversation across two models," with "a speech-to-speech model" owning the audio and "a reasoning model" doing "the thinking." Two of the practitioners we follow shared the tutorial link.
Shared on Bluesky by 2 AI experts
-
new piece featuring two of my favorite people: @zeu.dev and @craigsdennis.dev. they're showing how to combine voice models with phone calls and MMS to create really cool, multimodal experiences with @twilio.com and GPT-L…
View on Bluesky →
Originally reported by youtube.com
Read the original article →Original headline: How to Build a Voice AI Agent (GPT Live 1 + Twilio)