Alibaba released a real-time voice runtime that keeps agents talking while working

Voice assistants usually go silent and freeze while running a complex task—Alibaba's new runtime fixes that.

Alibaba's Qwen team open-sourced qwen-audio-agent, a real-time voice runtime for AI agents. Built on Node.js, it handles full-duplex continuous audio streaming so you can speak to an assistant naturally. While the conversation stays live, the framework hands off heavy work—like running code or calling tools—to background agents using the Agent Communication Protocol (ACP).

Why it matters: Most voice interfaces block the conversation the second an AI triggers a backend tool. This runtime keeps the audio line open while tasks execute in the background. You can ask for status updates, interrupt the agent, or tweak instructions mid-task without starting over.

Try it: Install it globally with npm install -g qwen-audio-agent (requires Node.js 22.22.2+). It offers a terminal TUI, a web interface, and a macOS desktop app with a floating visual orb. You can connect it to Alibaba's DashScope API or run it fully local using Hugging Face's speech-to-speech.

Finally, an AI voice assistant that doesn't put you on hold while doing real work.