Qwen Audio Agent is a real-time voice runtime for persistent agent dialogue, with frontend and optional backend agents, multiple installation methods, and extensive configuration. It supports OpenCode/OpenClaw backends and a gateway architecture for desktop and TUI interfaces.
What it is
Qwen Audio Agent is a real-time voice runtime that keeps agents talking, working, and present. It provides a frontend real-time voice interface with optional backend agents and gateway support for desktop and web/UIs.
How it works
The project supports multiple backend agents via AGENT_PROTOCOL (e.g., openclaw, opencode, qoder, acp). It includes a Gateway component, a desktop macOS app with a floating window, and a TUI/web UI for interaction. It enables continuous dialogue with back-and-forth, multi-turn conversations, and task-tracking where tasks can run via backend agents. Documentation references configuration for enabling specific backends, memory, and user profiles.
Getting started
Install methods include:
npm install -g qwen-audio-agent
npm install -g git+https://github.com/QwenAudio/qwen-audio-agent.git
From source:
git clone https://github.com/QwenAudio/qwen-audio-agent.git
cd qwen-audio-agent
npm install
npm run install:global
Upgrade commands:
npm install -g qwen-audio-agent@latest
npm install -g git+https://github.com/QwenAudio/qwen-audio-agent.git
Then:
qwenaudio config
DASHSCOPE_API_KEY=your-key
QWEN_AUDIO_REALTIME_MODEL=qwen-audio-3.0-realtime-plus
AGENT_PROTOCOL=openclaw
QWEN_AUDIO_AGENT_BACKEND_MODEL=qwen3.7-max
To start gateway and UI:
qwenaudio
qwenaudio tui
qwenaudio webui
For local Hugging Face speech-to-speech frontend:
pip install "speech-to-speech[paraformer]"
speech-to-speech \
--stt paraformer \
--llm_backend transformers \
--device cuda
speech-to-speech \
--stt paraformer \
--llm_backend mlx-lm \
--device mps
"} }```