-
Notifications
You must be signed in to change notification settings - Fork 198
Home
A realtime full-duplex voice runtime for AI coding agents — talk to Claude Code, Codex, OpenCode and more like a phone call.
qwen-audio-agent is an open-source voice frontend for coding agents. It listens, understands, and speaks in real time, delegating deep work to your backend agent through the open ACP protocol — then announces results back into the conversation.
- Full-duplex: speak while the agent works; interrupt anytime (barge-in)
- Task delegation: quick questions answered instantly, deep tasks routed to the backend agent
- Spoken results: long-running task outcomes are read back aloud
- Local wake word: on-device wake detection via sherpa-onnx
- Voice permission gates: destructive operations require spoken confirmation
User manual site (bilingual): https://qwenaudio.github.io/qwen-audio-agent/ — quick start, desktop/CLI/WebUI guides, and backend setup in one place.
Install: npm install -g qwen-audio-agent (Node.js 22+), then run qwenaudio config (DashScope API key + backend choice), then qwenaudio tui to start talking. Interfaces: macOS floating orb (custom skins since v1.7.0), TUI, and Web UI (English interface since v1.8.1).
11 backends: Qwen Code (default), Claude Code, Codex CLI, OpenCode, OpenClaw, Qoder, Hermes, Kimi CLI, CodeBuddy, DeepSeek Harness, and Pi — any ACP-capable agent plugs in. See Supported Backends.
- User manual: https://qwenaudio.github.io/qwen-audio-agent/
- Repository: https://github.com/QwenAudio/qwen-audio-agent
- Releases: https://github.com/QwenAudio/qwen-audio-agent/releases
- Discussions: https://github.com/QwenAudio/qwen-audio-agent/discussions