Grok Build v0.2.107 Changelog

This release is all about making the AI agent more flexible and persistent. It matters most to developers who want the AI to handle longer, more complex tasks without needing to be micromanaged, letting it pick up exactly where it left off—even if they switch computers.

What's new

• Smarter stop hooks: Normally, a "stop hook" tells the AI to shut down when it finishes a task. Now, it can feed advice back to the AI to keep it running instead of stopping.
• Rotating security tokens: If you use custom AI models, you can now set up a command that generates temporary, rotating passwords (tokens) to log in securely, rather than using a single static password.
• Better feedback details: When you leave feedback for the AI, it now records who wrote it, making it easier to track conversations and follow up.
• Cross-computer sessions: You can now save your session transcripts (the history of what the AI did) to external cloud storage like Amazon S3. This means you can start a session on one computer and resume it on another.
• Smoother auto mode: In "auto mode," if a safety filter blocks the AI, it now tells the AI *why* it was blocked and lets it try again. It only escalates the issue if the AI keeps getting blocked.
• Eager saving: You can now set the session storage to save after every single action (frame) instead of just at the end of a turn, so you lose less data if something crashes.

Improvements

• Faster recap summaries: When you leave the AI idle for a bit, the recap summaries it generates now load much faster by reusing cached (saved) information from the previous turn instead of starting from scratch.

Fixes

• Keyboard shortcut mix-ups: The Ctrl+B shortcut now correctly sends running commands to the background, and Ctrl+G properly opens the tasks pane.
• Login redirects: Pop-up login windows in the live preview now redirect you to the right place after you log in.
• Better Git awareness: When the AI starts up, it now sees all your file changes—including unstaged and untracked files—so it has a more accurate picture of your project.
• Randomized parameters: Sometimes the AI's tool settings get randomized for testing. The descriptions for those tools were getting messed up, but now they stay correct.
• Minimal mode fixes: In "minimal mode," the AI's full reasoning is now visible when you scroll up, and successful lookups are neatly collapsed into single lines so they don't clutter your screen.
• Stopping infinite loops: Empty commands (like true or an echo with no text) used to make the AI spin in circles. Now, it reminds the AI to stop and wait for background work to finish.

Breaking changes

No breaking changes in this release.