Leon van Zyl logs and labels GitHub issues with dependencies for every piece of work, then tells Claude Code to read the outstanding issues and implement them one by one.
Talk to Claude Code by voice while it builds, with Qwen Audio Agent
AICodeKing connects the open-source Qwen Audio Agent to Claude Code so he can keep a voice conversation going, interrupt answers and change requirements while a VoiceDesk app is built in the background.
AICodeKing · YouTube · published Oct 29, 2026 · Software & DevOps
▶ Watch on YouTubePrompt
Build a responsive issue tracker called VoiceDesk in the configured workspace. Use plain HTML, CSS, and JavaScript. Give it a dark theme, an issue list, and a new issue button. Let me add an issue, mark it complete, and filter completed issues. Save the data in local storage. Add basic validation and tell me how to run it locally. Keep all changes inside this project.
Visual breakdown
Qwen Audio Agent puts a voice front end on Claude Code: you keep talking, ask questions and change requirements while Claude builds in the background, then verify the real app instead of trusting the spoken 'done'.
The problem: Once Claude Code is working you can't easily add a forgotten requirement or ask a quick question without waiting for the whole coding task to finish.
How voice reaches the coding agent
- Voice in the browserBrowser UI on port 3101 with microphone; transcript and background tasks stay visible
- Voice front endQwen audio model via DashScope answers general questions and keeps talking
- Local gatewayConnects the voice side to the backend and brings task results back into the conversation
- Claude Code backendReached through an ACP adapter; does the real project work in the configured workspaceDropped: Spoken acknowledgement is not proof the file changed
- Checked appVoiceDesk opened and tested by hand, diff and regression check inspected
InputsSetup / glueAI agentResult
The tests he judges it by
- Did it understand the spoken request?
- Could the conversation continue while the backend worked?
- Did the follow-up reach the project?
- Did the finished app pass the checks?
By the numbers
Set it up yourself
Command-line install plus the browser client, as shown in the video.
- Use Node 24 series (2415 shown) with npm 10 or newer
- Install the package: npm install -g [email protected] (exact command is in the video description)
- Run qn-audio config, then fill config.env: DashScope API key, realtime provider, the qn audio 3.0 realtime plus model
- Check Claude works on its own, run qn-audio install claude and qn-audio setup --backend claude; set agent protocol to Claude
- Set the workspace field to the full absolute path of a separate project folder
- Start the gateway, then the web UI in a second terminal, allow the microphone and test with a short spoken request
After any config change, stop and restart the gateway.
Gotchas from the video
- Opening a terminal in a folder doesn't make it the agent's workspace; check the configured path.
- Interrupting the spoken answer does not cancel a task the backend already accepted; use the task cancel control, which doesn't undo edits already made.
- A mid-build change may become follow-up work rather than change the current turn; inspect the task state and final files.
- Use text for exact paths, code and errors; voice is for explaining and describing bugs.
- Open source isn't free to run: DashScope voice and Claude usage are billed separately.
Chapters
0:00The problem: talking while Claude Code works0:46What Qwen Audio Agent is1:48Node version and install2:40Voice provider config3:06Claude backend setup3:57Workspace folder and gateway5:12Building VoiceDesk by voice6:29Asking questions mid-build, interrupting7:44Changing a requirement mid-build9:00Bug report: spaces-only title10:43Cancelling a task11:33Costs and when to use it
Distilled from the video’s captions
How to try it
- Open Qwen Audio Agent.
- Copy the prompt and fill in the [placeholders].
- Watch AICodeKing's video for the full walkthrough and setup.
Related use cases
Skill Leap AI gives Claude Code a prompt and the path to about 400 memory and knowledge files, and gets a 3D castle with rooms, a memory wall and a bookshelf of clickable books.
Igor from The AI Advantage points ChatGPT (or Codex/Claude) at the open-source Reverse Engineer Anything repo, inside a project with folder access, and asks it to make a version of an app he paid for. He recreated Screen Brush as AIA Ink and a Slido clone.
Income Stream Surfers shares a free Claude Code prompt that audits a repo, mocks up a cleaner UI for approval, regroups menus into 4-7 areas, and ships it behind a feature flag.