Skip to content
HUIAHUB

QwenAudio/qwen-audio-agent

qwen-audio-agent

A realtime voice runtime that keeps Agents talking, working, and present. Real-time Voice Runtime for AI Agents

Metrics updated Emerging
View on GitHub (opens in a new tab)Homepage / demo

GitHub metrics

Stars
2.9K
growth after 7 days of data
Forks
290
30-day growth
—
Open issues
14

The chart appears once the repository has been checked on two or more days.

What it does

qwen-audio-agent is a realtime voice runtime that sits in front of a coding or task agent. You talk to a voice model, longer jobs go to a backend agent while the conversation continues, and when a task finishes its result comes back into the same conversation.

The voice side is swappable. Cloud options listed in the README include Qwen Audio 3.0 Realtime, GPT-Live / OpenAI Realtime, Google Gemini Live and Doubao Seeduplex 3.0 Realtime, each needing that provider's API key. Hugging Face Speech-to-Speech and MiniCPM-o 4.5 can run locally. Backend agents connect mostly over ACP: the list covers Qwen Code, OpenCode, OpenClaw and Kimi Code, with Codex and Claude Code joining through an external ACP adapter.

You can use it from a WebUI, a terminal TUI or a desktop floating orb on macOS, Windows and Linux. The README also lists parallel background tasks you can query or cancel, cross-session memory, and scheduled reminders. The package is published on npm under Apache-2.0, a technical report is on arXiv, and the repository sits in the QwenAudio GitHub organization.

Why it matters

The problem the README describes is a voice agent that goes quiet while it looks something up or runs a tool. Here the realtime voice model answers what it can and hands longer work to a backend agent, so you can keep talking, ask about progress or cancel. It reuses that agent's model configuration, tools, MCP, Skills and authentication, which adds voice to an existing setup rather than replacing it.

Main features

  • Full-duplex realtime voice with interruption and multi-turn conversation
  • Swappable voice frontends, with cloud and local options
  • Backend agents connect over ACP, including Qwen Code, Codex and Claude Code
  • Parallel background tasks you can check on or cancel
  • Task results return to the active conversation for follow-up questions
  • WebUI, terminal TUI and a desktop floating orb for macOS, Windows and Linux
  • Long-term per-user memory across sessions
  • Installs and manages Agent Skills

Who it's for

  • Developers who use Claude Code, Codex, OpenCode or Qwen Code and want to drive them by voice
  • Teams prototyping voice assistants such as customer service or smart cockpit demos
  • Users willing to supply cloud API keys, or to run a local speech-to-speech service

Strengths

  • Works with many existing coding agents instead of shipping its own
  • The voice provider can change without touching the backend agent
  • A local voice path exists through Hugging Face Speech-to-Speech
  • Conversation and background tasks run in parallel

Limitations

  • Most voice frontends are cloud services with their own keys, pricing and quotas
  • The README marks some integrations as not yet fully verified
  • MiniCPM-o 4.5 has no tool calling, and AI Passport is half-duplex only
  • Linux desktop builds (AppImage and deb) are unsigned

Getting started

The README recommends a global npm install. Desktop apps are built from a source checkout with one npm script per platform (desktop:build:local is the macOS build).

npm install -g qwen-audio-agent
npm run desktop:build:local
npm run desktop:build:win
npm run desktop:build:linux

Requirements: Node.js 22.22.2+ or 24.15.0+ and npm 10+; a DashScope/Bailian API key for the default voice frontend, or another provider's key or a local speech service.

Always follow the official instructions in the README on GitHub. This overview is our summary, not a copy.

Sources