Google lets Gemini run repetitive Mac chores in background window
Gemini on Mac can now navigate desktop apps and websites to finish multi-step tasks.

· Last verified: 20 Nov 2026 (Google blog, 9to5Google, TechCrunch)

Summary
- Gemini on Mac can now navigate desktop apps and websites to finish multi-step tasks.
- The agent runs in an isolated background window, pausing for user approval on key actions.
- Google enters the desktop-agent race already contested by OpenAI, Anthropic and Meta.
The latest
Gemini can now operate a Mac on a user's behalf, navigating desktop applications, local files and websites in a separate background window while the main screen and cursor stay free. Google announced the computer-use capability on November 19, 2026, in a post by Michael Friedman, group product manager on the Gemini app. It is launching in English in the United States only.
Details
- How it works: Google said Gemini operates inside an isolated background window with a live preview streaming into the chat. Users pick which applications the agent can reach, watch each step as it happens, pause it at any moment, and approve major actions before they execute.
- The guardrails: According to 9to5Google, Gemini can be halted with a single click at any point. Google stops the agent at login screens and payment steps, and requires explicit confirmation for sensitive actions rather than proceeding on its own.
- How it is triggered: 9to5Google reported the feature is invoked by adding @Computer use to a prompt. Under Customize, then Connectors, new desktop integrations including Blender and Autodesk Fusion join existing connections to Git and Obsidian.
- The use cases: Google's examples include recoloring batches of images, preparing 3D scenes, extracting weights, dimensions and addresses from local PDF files to request shipping quotes, and pulling metrics and screenshots from dashboards into a Google Slides deck.
- More examples: The company also described analyzing photographs of clothing in a local folder and drafting resale listings from them. TechCrunch listed filling out forms, repeating one task across a batch of files, and sorting files into folders by criteria.
- Who can use it: The rollout requires a Mac with Apple Silicon running macOS Sequoia 15.0 or later. It is open to personal Google accounts for users aged 18 and over, and to subscribers of Google AI Plus, Pro and Ultra, with downloads at gemini.google/mac.
- The competition: TechCrunch, in a November 19, 2026 report by Sarah Perez, noted that OpenAI, Anthropic and Meta already offer agentic features inside their own desktop applications, putting Google into a contest it entered after its largest rivals.
- What Google withheld: The blog post does not name the model powering the feature and lists no pricing. Google has not said when the capability will expand to other markets, other languages, or platforms beyond Mac.
Between the lines
The design choices read as a bet on trust rather than autonomy. An isolated window, a live preview, per-app permissions, a one-click stop and hard halts at logins and payments all hand control back to the user at the moments where an agent operating a personal machine could do the most damage — and where the slowest adoption barrier sits.
What's next
Watch for Google to name the underlying model, extend the feature beyond English and the United States, and add platforms past Apple Silicon Macs. Also watch which desktop connectors join Blender, Autodesk Fusion, Git and Obsidian.