Summary
On October 15, 2026, Cloudflare made @cf/daily-co/phonellm-mini-alpha-2 available on Workers AI, a voice-agent model from the Pipecat team at Daily built by post-training NVIDIA Nemotron 3 Nano 30B on phone and customer-support scenarios for better instruction following and tool calling without reasoning mode. Cloudflare also shipped a Pipecat voice-agent example running in a Durable Object over WebRTC.
What changed
Cloudflare added PhoneLLM Mini (@cf/daily-co/phonellm-mini-alpha-2) to Workers AI, a Nemotron 3 Nano 30B model post-trained by Pipecat/Daily for voice and realtime agents with improved instruction following and tool calling (recommended with temperature 0 and thinking disabled), plus a sample Pipecat voice agent running in a Durable Object over WebRTC and defaulting to DeepSeek V4 Flash. Python Workers are now generally available.
Why it matters
Hosting a purpose-built voice-agent model alongside a WebRTC Durable Object example and GA Python Workers lets developers build low-latency phone and support agents entirely on Cloudflare's edge, strengthening its position as a full-stack platform for realtime agentic applications.
Evidence excerpt
@cf/daily-co/phonellm-mini-alpha-2 is a model from the Pipecat team at Daily, built for voice agents and realtime applications. A Pipecat voice agent example runs in a Durable Object over WebRTC, defaulting to DeepSeek V4 Flash.