Gemini CLI v0.32.1: Seamless Autocomplete, Folder Trust and UI Polish

[srithreepo]

Here is Gemini CLI’s weekly update for v0.32.1 🚀

✨Gemini 3.1: While we work on bringing in Gemini 3.1 Flash-lite support both more natively and with our sub-agents, routing and summarization you should already be able to use Gemini 3.1 Flash-lite when using Gemini CLI with the -m option when using your API Keys

⌨️ Interactive Shell Autocompletion: You can now use readline-style Tab autocompletion in shell mode! This includes robust file and directory pathing (even for dotfiles and paths with spaces), plus context-sensitive suggestions for developer tools like git and npm ( #20082, @mrpmohiburrahman ).

🛡️ Massive Security & Policy Upgrades:

• Folder Trust & Warnings: When installing extensions or trusting a folder, the CLI now runs a "Discovery" phase to proactively flag security-sensitive settings (like disabled sandboxes) before you grant access ( Folder trust dialog is unclear #15010, @galz10).
• Deceptive URL Detection: Added protections against homograph attacks and deceptive Unicode characters (e.g., hidden BiDi/zero-width characters) to keep your terminal and tool confirmations secure (security: strip deceptive Unicode characters from terminal output #19026, @ehedlund).

🗺️ Plan Mode (Experimental) Polish:

• Automatic Model Switching: Plan mode is now smarter about routing. It automatically uses high-reasoning Pro models during the planning phase and switches to high-speed Flash models for execution (Support model switching for Plan Mode #19454, @jerop).
• External Editor Support: You can now open and modify your generated plan in your preferred external editor simply by pressing Ctrl+X during the approval phase (Support opening and modifying current plan in editor #17721, @Adib234).
• Read-Only Enforcement: Enforced strict read-only constraints during the planning phase to prevent unintended code modifications before the plan is approved ([Agent] Reduce policy failures due to implicit constraints during planning #19437, [Tools] fix plan mode failing to write to plan file in Windows  #19648, @mattKorwel, @jerop, @Adib234).

🧠 Agents & Pipelining:

• Tool Pipelining: Optimized multi-turn operations to eliminate unnecessary steps. For example, grep_search now automatically returns file content for small result sets, and write_file returns a diff, saving the agent from needing to do follow-up verification reads ([Context Mgmt] Investigate pipelining of tool calls to reduce token consumption and scenario duration #19444, @gundermanc).

👁️ UI/UX Enhancements:

• 📊 Compact Context Display: The context window indicator has been streamlined for a cleaner UI. Instead of showing the percentage of context left, it now shows a compact display of the percentage of context used (feat(cli): invert context window display to show usage #20071, @keithguerin ).
• macOS Notifications: Added opt-in macOS native notifications (general.enableMacOsNotifications) for action-required states (like tool approvals) and task completions (Implement Desktop Notification for User Permission Prompts #14696, @LyalinDotCom).
• Customizable Loading Phrases: You can now configure the loadingPhrases setting to tips, witty, all, or off for a quieter or more informative wait state (Remove unprofessional witty phrases #14942, @LyalinDotCom).

⏳ MCP Progress Bars: Long-running MCP tool calls now support visual progress bars and throttling, giving you real-time percentage and message feedback right in the terminal UI (#16934, @jasmeetsb).

Small features, Polish, Reliability & Bug Fixes: A large amount of changes, smaller features, UI updates, reliability and bug fixes + general polish made it in this week!

[raky291]

Thank you so much for the great work you have done with Gemini CLI.

[HenkPoley]

I see none of the Gemini 3.1 models in Gemini CLI, when logged in through OAuth with my Google Account.

gemini-3-pro-preview is going to be turned off soon. Aren't you forgetting some people? (This is probably an API/server-side thing.)

Edit: One should know that the 'Google AI Plus (200 GB)' plan is seen as equivalent to Free, by Gemini CLI. That is OK, Google should set their price. But the tool and the website should be more upfront about that.

[garrylea]

When will the replace function be improved? It consistently fails when updating larger code files. Furthermore, when I resort to a full-file update, the model often hallucinated and completely mangles the original logic, making it unrecognizable.

[aniruddhaadak80]

From my point of view, weekly updates like this become much more useful when they also include one short section on known rough edges after the release. People reading this thread are often trying to decide whether to upgrade immediately or wait. A compact list of rollout caveats and current limitations would make these release posts much more actionable.

[mrpmohiburrahman]

ok, but why posting same comments 3 times!?