Qwen Code v0.19.2 Changelog

This release is all about making the AI coding assistant more flexible and safer to use. It matters a lot to developers who use this tool to write code, especially those who want to use voice commands, work with images, or connect outside tools to the AI without worrying about accidental damage.

What's new

• Voice dictation in the Web Shell: You can now speak to the AI using your microphone in the web interface, and it will type out your words.
• Smarter voice transcripts: Before inserting your spoken words into the chat, the AI quickly cleans them up using a fast model so there are fewer weird mistakes.
• Image reading for text-only models: If you are using an AI model that can't normally "see" images, the tool will now automatically translate the picture into a text description so the AI can still understand it.
• Safety guards for auto mode: When the AI is running automatically, it now has strict blocks to stop it from running destructive commands (like deleting important files).
• Workspace permissions rules: There is a new system to set specific rules for what the AI is allowed to do inside your project workspace.
• Better external tool matching: If you type the "@" symbol to mention an outside tool (using the MCP standard), the system is now smarter at finding what you mean and showing you the full reference.
• Read external resources: Added a new tool that lets the AI read data from connected external MCP resources.
• Remote status check: Added a way for other programs to check the status of the language server (the tool that checks your code for errors) from afar.

Improvements

• Faster startup: The background program (daemon) that helps run the web shell now starts up much faster.
• Cleaner chat UI: The chat interface in the web shell was restructured to look and feel better.
• Simpler icons: Replaced emoji icons used for "thinking" and "summary" with simple text symbols, which makes the interface look cleaner and load faster.
• Better testing: The automated tests that check the code (CI) were reorganized to be faster and more reliable, meaning future updates will roll out smoother.

Fixes

• Chat view disappearing: Fixed a bug where the chat view would sometimes vanish from the VS Code sidebar.
• Image paste bug: If you pasted a file path for an image into the chat, it wouldn't attach properly. Now it correctly adds the image as an attachment.
• Duplicate names: If you had two AI models with the exact same name, it caused confusion. The system now makes sure they are clearly told apart.
• Annoying backgrounds: Removed weird background colors that were messing up the input box and your chat messages.
• Mouse glitches: Fixed unstable mouse interactions in the terminal view so clicking works smoothly.
• File position errors: Fixed a bug in VS Code where opening a file could jump to the wrong spot or crash.
• Settings loading: If the app couldn't find its main settings file, it now correctly falls back to your personal environment files.
• Strict number settings: Fixed settings for "stop hooks" and "microcompaction" (memory saving) so they only accept whole numbers, preventing crashes.
• Lost custom models: Fixed an issue where your custom chosen AI models would be forgotten if you went back into the login setup wizard.
• Stale connections: The system now rejects old, frozen prompt clients and can heal broken login IDs automatically, so you don't get stuck.
• MCP transport mapping: Fixed how Claude MCP servers connect, making sure the right communication method is used when importing them.
• Context token counting: Fixed the counter that tracks how many words/tokens you have used in a chat so it pulls the right number from your specific session.
• Fork limits: Added a cap on how many times an AI task can "fork" (split into a new branch of thought) and fixed it so it properly asks for permission before forking.

Breaking changes

No breaking changes in this release.