Gemini CLI v0.41.0 Changelog

This update is a big deal for anyone who uses the Gemini command-line tool (CLI) to talk to AI directly from their computer's terminal. It focuses on making the app much safer to run, adding exciting new ways to interact—like using your voice or trying out Google's experimental Gemma 4 models—and making the app start up much faster.

What's new

• Real-Time Voice Mode: You can now talk to the CLI using your voice, with options to process the audio either on your computer (locally) or in the cloud.
• Gemma 4 Support: You can now experiment with Google's next-generation Gemma 4 lightweight AI models.
• Auto-Memory Scratchpad: The tool can now remember skills and information it learns about you across different sessions so you don't have to start from scratch every time.
• Manual Session IDs: You can now manually set a unique ID (UUID) for your chat session, making it easier to organize and resume specific past conversations.

Improvements

• Faster Startup: The app now loads background data (like your usage limits and experimental features) asynchronously—meaning it does this in the background so the app opens much quicker.
• Stronger Security: Added "workspace trust" (making sure a folder is safe before running code in it) and better checks on terminal commands to prevent unauthorized actions.
• Better Error Messages: If your internet drops (network timeouts) or you run into folder errors, the app now explains what went wrong in plain English instead of showing confusing computer codes.
• New Learning Resources: Added a link to a Gemini CLI training course in the documentation to help you learn how to use the tool like a pro.

Fixes

• Windows Backspace Bug: Fixed an issue where the backspace key stopped working properly for Windows users in the user interface.
• Hard Crashes: Stopped a severe crash that happened when the app got stuck in an endless loop.
• Output Redirection: Fixed a bug so you can now easily save the CLI's output directly into a text file on your computer.
• YOLO Mode Safety: In "YOLO mode" (where the AI runs commands automatically without asking you first), the app will now safely shut down and block commands if it can't verify they are safe, rather than risking running something dangerous.
• Temporary Network Glitches: Fixed an issue where temporary internet hiccups would completely stop the program; it now knows to ignore these quick glitches and keep running.

Breaking changes

No breaking changes in this release.