Qwen Code v0.23.0 Changelog

This release is all about giving you more control and better visibility over what the AI is doing. It matters most to developers and team managers who use Qwen Code to manage complex software projects, especially through the Web Shell. It introduces better ways to handle background tasks, keep track of GitHub pull requests, and ensure the AI communicates exactly the way you want it to.

What's new

• Smarter Branch Picker: When you're about to update or commit code, the menu now shows hints (like "↓3" meaning you are 3 updates behind) so you know exactly what will happen before you click.
• Goal Proposals: The AI can now suggest a "Goal" for what it's about to do, and a pop-up dialog lets you approve it first.
• Custom Output Styles: You can now choose how the AI talks to you (like super brief or very detailed) using a simple setting or command flag. It will also remind itself every turn to stick to that style!
• Web Shell Upgrades: You can now start standalone chats, scan a QR code to quickly pair your phone or device, and easily copy error messages right from the browser terminal.
• GitHub PR Links: If you are working on a GitHub Pull Request, the Web Shell now shows you exactly which issues that PR will fix, and lets you search your past sessions by issue number.
• Workflow Controls: You can now pause, resume, retry, or cancel AI workflows (a series of automated steps) right from the app.
• Private Network Extensions: If you work for a company, you can now install Quen extensions from private, internal Git servers, not just public ones.
• Scheduled Tasks: You can set up tasks to run on a schedule in the Web Shell, and each run gets its own fresh, timestamped chat log so you can review what happened.
• Voice Daemon: A new standalone package for voice features (qwen-live) has been added for hands-free interaction.
• Better Memory Control: The AI's memory is now strictly divided by project and user, so it won't accidentally mix up rules or facts between different projects.
• Auto-assigning PRs: New Pull Requests are now automatically assigned to exactly one maintainer (the person in charge of that code area) to make sure nothing gets ignored.

Improvements

• Cleaner Session Tables: The menu showing your past chats is now much easier to read, with columns that adjust to your screen and let you sort and filter.
• Smarter Tool Choices: When the AI decides to use an external tool, it now looks at the actual arguments and server notes, not just the tool's name, helping it make better choices.
• Handling Messy Code: If you try to update your project but have uncommitted changes lying around, the Web Shell will now ask if you want to "stash" (temporarily hide) or discard them, instead of just blocking you.
• Hot-Reloading Providers: If you change which AI model you want to use, the app now applies it instantly without needing to restart your chat session.
• Terminal Foundation: The developers are building a new, modern terminal interface (called OpenTUI). This update adds the underlying pieces for themes, accessibility, and better text rendering, paving the way for a smoother experience soon.
• Channel Customization: If you use channels (like DingTalk) to talk to the AI, you can now customize the emojis used to show when a task starts and finishes, and error messages hide sensitive internal data.

Fixes

• Compression Errors: Previously, if the system failed to compress data (which it does to save memory), the error wasn't clear. Now, it specifically tells you when the compression API fails.
• Web Shell Buttons: On certain servers that lack pop-up directory pickers (like zenity or osascript), the "Browse" button would show up and do nothing. It is now properly hidden.
• DingTalk Privacy: When sending failure messages through DingTalk, the app used to leak internal error data. That is now safely hidden from users.

Breaking changes

No breaking changes in this release.