Mistral Vibe v2.24.1 Changelog

This update is all about making the AI coding assistant feel faster and more reliable. It matters a lot to developers who use this tool daily, as it smooths out rough edges, stops the app from crashing, and adds smart shortcuts to help you get your work done with less friction.

What's new

• Ghost-text autocomplete: The AI can now predict and finish your sentences right in the middle of what you're typing, kind of like how your phone suggests the next word.
• Quick-start commands: You can now start a brand-new chat (/clear or /new) and immediately give it your first question or prompt at the same time. It also shows you a handy ID number so you can jump back to that exact chat later.
• Smarter memory: When the app "compacts" a really long conversation to save memory, it now safely preserves that summarized history instead of losing track of what was said.
• App monitoring: Added support for OpenTelemetry. This is a tool that lets developers track how the app is performing behind the scenes to make sure everything runs smoothly.

Improvements

• Renamed default agent: The main AI assistant you chat with is now just called ask, and it's set up by default to accept the code edits it suggests.
• Better file organization: When the app automatically creates a "worktree" (a temporary copy of your project files so the AI can safely experiment), it now names that copy based on which AI model it's using.
• Faster chat recovery: When you jump back into an old conversation, it loads much faster now. It does this by waiting to load background "sub-agents" (helper AIs) until it actually needs them.

Fixes

• Better error messages: If a background server fails to start, or if the app messes up transcribing your voice to text, it will now actually tell you instead of failing silently.
• Visible text selection: When you highlight text to copy or delete it in the chat box, you can actually see it highlighted now. (Previously, it was invisible!)
• Smarter AI connections: If the AI provider gets cut off mid-sentence (an incomplete stream), the app now detects it and tries again. It also replays the AI's "thinking process" exactly as it happened.
• Login fixes: The app now automatically refreshes expired security passes (OAuth tokens) for external servers before they cause errors.
• UI tweaks: Switching between light and dark themes is now instant. Also, settings menus no longer cut off their titles when the text wraps to a new line.
• Crash prevention: Fixed a bug that crashed the app if you tried to "undo" pasting a giant block of text into the chat box.
• Security and folder fixes: The app no longer accidentally treats your entire computer's Home folder as a project. It also fixed a weird bug on Windows where using Git Bash accidentally created fake C:\c\... folders.
• Clearer settings: If you accidentally set the app to use an AI model that doesn't exist, it now explains why it's falling back to a default one instead of just confusing you.
• Tighter security: Fixed a loophole where the chat could bypass rules meant to stop it from reading sensitive files. Also, telling the app to trust your current project folder no longer accidentally makes it untrust a hidden .vibe folder inside it.
• Path formatting: When showing you file locations, the app now uses standard forward slashes (like /) instead of mixing them up, making them easier to read.

Breaking changes

No breaking changes in this release.