Mistral Vibe v2.19.1 Changelog

This update is all about giving you better control over the AI's tools, speeding up how fast the app starts, and fixing annoying glitches. It matters most to developers and power users who use this software to code, because it makes complex tasks smoother and much less frustrating.

What's new

• Test out a new bash tool: There's a new, experimental feature that lets the AI run and track terminal (command-line) sessions.
• Turn off specific tools: You can now use a new command-line flag (--disabled-tools) to easily switch off certain AI tools for a specific task.
• Custom tool descriptions: You can now write your own custom instructions for how specific tools should behave, using simple text files.
• Fresh start after planning: There's a new setting to clear the AI's short-term memory (context) once you accept a plan it made, so it starts working with a clean slate.
• Better file reading: When the AI reads a file, it will now also show you any special AGENTS.md instruction files it finds in that folder.

Improvements

• Smarter memory management: When the AI's memory gets too full, it now creates stricter, more focused summaries of the chat so it doesn't forget the important stuff.
• Faster startup: The app now loads much quicker by waiting to load heavy background features until they are actually needed.
• Smoother background updates: The panel that connects to outside tools (MCP) now refreshes quietly in the background without interrupting you.
• Less flickering: The loading spinner that appears when switching AI agents is now "debounced" (meaning it waits a split second before showing up, so it doesn't flicker if the switch is instant).
• Better privacy: If the app crashes and sends an error report, it now automatically scrubs out your personal info and private computer file paths.
• Smarter network settings: Web requests now better understand advanced network rules (like CIDR notation) for when you need to bypass a proxy server.
• Faster history loading: When you resume an old chat, it now loads only when you actually look at it, which speeds things up.
• Safer model fallback: If the app gets confused about which AI model it's supposed to use, it now automatically falls back to the first available one instead of breaking.

Fixes

• Rewind no longer clones: If you "rewind" the chat to undo a mistake, it now happens right in the same chat instead of creating a confusing, split-off copy of the session.
• Cleaner error messages: Error messages no longer include confusing web links from a coding library called Pydantic.
• Fixed text selection scrolling: You can now scroll your text properly while highlighting words at the very bottom of the chat window (previously, it got stuck).
• Better click highlighting: When you click and drag to highlight text, it now neatly snaps to whole words and paragraphs instead of cutting them off awkwardly.
• Fixed a crash: Fixed a bug that crashed the app if the screen-recording interface was accidentally started twice.
• Fixed Git text errors: Fixed an issue where text output from Git (the version control system) showed up as garbled characters instead of normal text.
• Fixed UI glitches: Corrected the placement of the symbol that shows text is cut off, and removed a duplicated web link.

Breaking changes

No breaking changes in this release.