Google tests Gemini reasoning mode with voice control and app settings

Google just shipped a Gemini desktop update that makes talking to your computer feel less like a sci-fi novelty and more like an actual workflow improvement. Version 1.88 of the Gemini macOS app introduces voice input that works across any active window, a reasoning mode that can analyze on-screen content, and granular controls that let users decide exactly how much power to hand over to the AI.

The update, which began rolling out on July 29, represents Google’s clearest move yet toward turning Gemini from a chatbot you visit into an assistant that lives inside your operating system.

How it works

The core mechanic is simple: long-press the Fn key, start talking, and Gemini transcribes your speech directly into whatever app you’re using. The system automatically strips out filler words and handles corrections on the fly.

Users can also set up custom keyboard shortcuts if the Fn key doesn’t suit their workflow. During active voice sessions, a floating waveform indicator provides visual confirmation that Gemini is listening.

The more interesting piece is what Google calls the “reasoning” mode. When enabled, Gemini can analyze whatever’s on your screen, whether that’s a document, an image, a spreadsheet, or a file, and perform complex tasks based on that context. Summarization, rewriting, information extraction: the kind of work that typically involves copying content from one app, pasting it into Gemini, waiting for a response, then copying the output back.

Privacy controls and the opt-in model

The screen-context reasoning mode is disabled by default. Users have to manually toggle it on in the app settings, which means Gemini won’t start analyzing your screen content unless you’ve explicitly granted permission.

This is a deliberate design choice that reflects the “agentic” AI philosophy Google has been promoting. The idea is to build AI that doesn’t just answer questions but takes action on your behalf, while still keeping humans in the consent loop.

From I/O preview to shipped product

Google previewed elements of this functionality at Google I/O in May 2026, giving developers and the press a look at where the Gemini desktop experience was headed. The July release translates those demos into production software that actual users can install and rely on daily.

The initial rollout is English-only, with global availability. Multi-language support is planned but not yet on a public timeline.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.