Grok Build v1.0.1 Changelog

This update is all about making the AI assistant faster, smarter, and way less likely to freeze up. It matters to anyone who pushes the software to its limits—like coding in massive projects or having the AI juggle multiple complex tasks at once—because it fixes a lot of the frustrating slowdowns and glitches that used to happen during heavy use.

What's new

• Smarter task management: When the AI spawns "subagents" (mini-helpers to do side tasks), it now puts them in a line instead of launching them all at once. This stops it from crashing by using up too many system resources.
• Disk usage check: A new command (grok du) shows you exactly how much storage space your AI history and saved work are taking up on your computer.
• Safer tool rules: Tools now have to report if they only read data. This lets the AI safely restrict what mini-helpers are allowed to do.
• Custom skills: You can now limit which special "skills" the AI is allowed to use in certain sandboxed (safe, isolated) workspaces.
• Easier session renaming: When you rename a chat session from the dashboard, it now shows the current name so you can easily edit it, keeps your custom names across different computers, and limits names to 100 characters. You can also use /rename --auto to let the AI name the chat for you again.
• Better info pop-ups: Commands that show your usage, session info, or context now open in a neat, tabbed pop-up window instead of cluttering your chat with text.
• Easier debugging: When you export a "trace" (a log of what the AI did to fix a problem), it now includes memory files to make troubleshooting easier.
• Upgraded video generation: Making videos from reference images now supports preset voices, single images, 1–15 second durations, and different screen shapes (like 4:3 or 3:4).

Improvements

• Huge speed boosts for big projects: If you use the AI in large coding projects, checking your Git status or looking through deep project history will no longer hog your computer's memory or freeze up the app.
• Faster session loading: Opening up large, long chat histories is now significantly faster. The screen also won't show an incomplete chat while it's loading.
• Better background management: Searching through your history no longer leaves hidden background processes running, which used to eat up memory over time.

Fixes

• Folder permissions fixed: Setting folder permissions ending in / now correctly grants access to the whole folder, instead of accidentally creating a weird folder literally named .
• Apple Silicon fix: On newer Macs, the app now correctly installs the native version even if you are using an older, translated terminal.
• Menu fixes: Custom skills that share names with built-in commands now show up properly in the menu. Also, annoying plugin suggestions no longer flash on the screen while you are typing.
• Language and input fixes: Chat summaries are now written in the same language you are chatting in. Pressing the "Home" or "End" keys now moves your cursor to the start or end of a typed line, even if your text is wrapped.
• No more freezing: The AI no longer hangs forever if a tool times out, if a background process gets stuck, or if a session tries to load a file that blocks the system. Also, deleting a session now safely waits for any running mini-helpers to finish before wiping the history.
• Smoother steering: If you type a message while the AI is already working (by pressing Enter twice or Ctrl+Enter), the AI now correctly understands you are adding mid-turn thoughts.
• Video error messages: Video tools now give you a clear error message about storage rules instead of just silently failing and disappearing.
• Cancel button fixed: Pressing Escape on the cancel-turn panel actually just closes the panel and lets the AI keep working, instead of doing something confusing.
• Other workflow fixes: Sending a message while the AI is working on a goal no longer cancels the goal. Headless sessions (running without a visual interface) now wait properly for servers. Non-interactive sessions no longer crash if the AI asks for your approval. Git worktree sessions now correctly show their branch name and keep their status when you switch folders or open the dashboard.

Breaking changes

• The /rewind command (which undoes recent steps) now only erases the chat history, not your actual files, and it will ask you to confirm before doing anything.
• "Managed MCP servers" (external tools the AI can connect to) can now only be found and used through the official gateway catalog, rather than from anywhere.