Qwen Code v0.21.7 Changelog

This release is all about making the coding experience smoother and less frustrating for developers. It matters most to people who use Qwen Code to review code or manage long, complex tasks, giving them more power, better speed, and fewer weird glitches to worry about.

What's new

• Images in the terminal: If you use a compatible terminal (like Kitty, Ghostty, or chafa), the AI can now display images directly in the command line. You no longer have to open a separate window to see what the model is showing you.
• Custom code review rules: You can now create a special settings file (.qwen/review-context.json) to teach the AI exactly how you want it to review your project.
• Private voice transcription: For larger, managed setups, admins can now route voice-to-text features through private, secure networks instead of the public internet.
• Better monitoring: Added behind-the-scenes tracking for data streams. This helps developers spot issues if a user's internet is too slow to keep up with the AI's responses.

Improvements

• Faster code reviews: The system used to take little pauses between checking your code and verifying the fixes. Those delays were removed, so the whole review process is faster.
• Cleaner chat updates: If you use DingTalk (a work chat app), the AI now updates its progress in a single, neat message bubble instead of spamming a bunch of separate cards.

Fixes

• No more 50-turn limit: Previously, if you gave the AI a big goal, it would just stop working after 50 back-and-forth steps. That limit is gone, so long tasks can finally finish.
• Fixed frozen GitHub reviews: Sometimes, asking GitHub to review code would cause the system to hang silently forever. It now has a built-in timer to stop the stream if it takes too long, preventing these freezes.
• Safer file access: Fixed a security glitch where shortcuts (symbolic links) could trick the system into reading files outside your project folder. It now checks where the shortcut actually points before allowing access.
• Fixed app signing: The code that proves the desktop app is safe and officially made by the developers was failing because of a naming mismatch. The names are now fixed.
• Instant branch updates: When you switch code branches in the Web Shell, the screen now updates the branch name immediately instead of making you wait.
• Refresh button fixed: You can now hit refresh in your browser on a Web Shell page without getting an error, while keeping the actual code commands safe and secure.
• Fixed missing code in reviews: The review tool was accidentally leaving out a core piece of code when it ran, causing checks to fail. It now includes everything it needs.
• Clearer sandbox errors: When the system tries to run code in a safe, isolated environment (a "sandbox") and can't find one, it now gives you a clear error message instead of just failing mysteriously.
• Stopped overlapping auto-fixes: Fixed a timing glitch where the AI might try to scan and fix the same code twice at the exact same time, which caused crashes.
• Fixed language settings: You can now change the interface language locally without the command accidentally ignoring your project settings.
• Kept files in their lane: Fixed an issue where files, downloads, and reports from one project workspace could get mixed up or locked after moving to a different workspace.

Breaking changes

No breaking changes in this release.