Qwen Code v0.19.4 Changelog

This release is all about making the AI assistant more flexible and less frustrating to use. It matters to anyone who uses Qwen Code to build software, especially if you use the web interface, rely on background agents to do automated work, or need to connect the tool to custom, private servers. The developers have added highly requested features like voice dictation and safe mode, while spending a lot of time squashing bugs that caused the AI to freeze up, glitch the screen, or lose track of what it was doing.

What's new

• Voice dictation: You can now speak to the desktop app instead of typing your prompts.
• Safe mode: A new --safe-mode flag turns off all custom settings and plugins. It’s perfect for figuring out if a bug is caused by your custom setup or the app itself.
• Skip security checks: A new --insecure flag lets you connect to private, self-signed servers (servers without official web security certificates) without the app blocking you.
• Better interruptions: If the AI is talking and you hit the Esc key to stop it, or if you type a new prompt while it's working, the app handles it much more smoothly now. It will even queue up your questions to answer when it finishes its current thought.
• Mobile web layout: The web version now has a sidebar drawer and a browser tab icon, making it feel much more like a real app on your phone.
• Quick settings: You can now type /config key=value right in the chat to instantly change settings without digging through menus.
• Multiplayer agents: The developers added the groundwork for "channels," allowing multiple AI agents to live and work together in shared spaces.
• Smarter memory limits: You can now use wildcard patterns (like *) to easily allow or block certain tools from accessing the AI's memory.

Improvements

• Smoother task resumption: If the AI gets interrupted while working on a task, it can now pick up right where it left off without you having to type "continue."
• Better web shell: The web interface now shows cleaner file differences (diffs) when the AI edits code, and gives you better follow-up suggestions.
• Mouse support: You can now click and hover over things in the app's alternate-screen mode (the full-screen terminal view).
• Easier model switching: You can quickly override the AI model you're using for just one specific chat without changing your global settings.
• Cleaner interface: The app replaced standard emojis with text symbols and changed a specific icon to stop characters from glitching and overlapping on certain computer screens.

Fixes

• Infinite loops fixed: The AI used to get stuck repeatedly trying to use a broken tool or inspecting the same shell command over and over. These loops have been stopped.
• Memory recall fixed: The app was trying to remember things even when the auto-memory feature was turned off. It now only searches its memory when you actually want it to.
• Screen glitch fixes: Fixed issues where the "thought viewer" (how the AI shows its reasoning) would cut off text, leave weird gaps, or scroll choppily.
• Windows path fix: Fixed a bug where the app couldn't understand file paths that used the ~ symbol (which means "home directory") on Windows.
• Accurate token counts: Subagents (smaller AIs that help the main AI) were showing the wrong number of output tokens. This is fixed so you get accurate info.
• Crash prevention: Fixed an "Out of Memory" issue where the app tried to download too much history at once in certain situations.
• Cleaned up AI thoughts: Fixed a bug where the AI's internal "thinking" text would accidentally leak into the final results when using certain models like GLM.

Breaking changes

No breaking changes in this release.