Google Gemini Arrives on Mac as a Native Swift App With Screen Sharing

Photo by Dmitry Chernyshov on Unsplash

Google shipped a dedicated Gemini app for macOS on April 15, 2026, closing a gap that had left it as the only one of the three major AI assistants without a native Mac client. The download lives at gemini.google/mac, the app is free, and it is built entirely in Swift rather than wrapped around a web view. That last detail is the one Google wants people to notice.

The launch was announced on the Google Workspace Updates blog, which describes the release as a native desktop experience open to personal Google accounts, Workspace business and education customers, and Workspace Individual subscribers. Two global shortcuts handle most interaction: Option + Space opens a compact chat surface that floats over whatever you're working on, and Option + Shift + Space opens the full window. Both are configurable in the app's settings.

Google VP Josh Woodward used the launch to make a point about how the app was built. According to 9to5Mac, Woodward posted that the client is "100% native Swift" and that the team "built 100+ features in less than 100 days," working with the agency Antigravity to go from prototype to shipping product in a matter of days. Sundar Pichai marked the moment on the same day, noting it was "the first time we're bringing the @Geminiapp to desktop."

That framing matters because it puts Gemini on the same architectural footing as its competitors. ChatGPT has had a Mac app since May 2024, and Claude shipped its own in October 2024, per 9to5Mac's recap citing MacRumors. Gemini is the last of the three to land on the Mac, but Google is arguing it arrived with the right foundation rather than a quick Electron port.

What screen sharing actually does

The headline feature is screen sharing inside the chat surface. Users can hand any active window to Gemini and ask questions about its contents, with the Workspace Updates post giving the example of pointing Gemini at a chart and asking, "What are the three biggest takeaways here?" The model reads what's on screen and responds in the same chat window.

This is where the native Swift build pays off in a practical way. Capturing window contents, handing them to a vision-capable model, and streaming a response is the kind of workflow that exposes the seams in a webview-wrapped app. A native client can negotiate macOS screen recording permissions directly, throttle capture intelligently, and integrate with the system permission prompts users already trust.

Google addresses the obvious privacy question at first launch. According to coverage from Fello AI, Google's own permissions disclosure states that screen content is not stored beyond the duration of each query. That is narrower than some users may want, since query-time processing still means the content leaves the device, but it rules out the worst-case scenario of persistent screen archives sitting on Google's servers.

Models, generation, and what runs inside the window

The app does not run a single Gemini model for everyone. Free users get Gemini 3.1 Pro, and subscribers on Google AI Pro or Ultra tiers get Gemini 3.1 Ultra, both with full vision support, according to Fello AI's review citing model documentation. Vision support is what makes the screen sharing feature work in the first place, so the free tier is not crippled in the way a generation gap might suggest.

Image generation runs through Nano Banana and video generation through Veo, both invoked directly inside the chat window rather than as separate destinations. Fello AI notes that Veo's video output is largely paywalled, which is consistent with how Google has positioned Veo across its other surfaces.

The Gemini release notes describe a parallel wave of changes that arrived with the desktop client. Chat-history import tools let users pull memories and chat archives out of competing AI apps and into Gemini, an unusually aggressive move that lowers the switching cost for someone who has accumulated months of context inside ChatGPT or Claude. The same release renames "Past chats" to "memories," aligning Gemini's terminology with how the rest of the industry talks about persistent context.

System requirements and where to get it

The download is exclusive to gemini.google/mac. There is no Mac App Store listing at launch, per the release notes, which also confirms macOS 15 Sequoia or later as the minimum operating system and free global availability.

Fello AI cites Dataconomy's hardware breakdown: macOS Sequoia 15 or later, at least 8 GB of RAM, 200 MB of free disk space, and a stable internet connection. None of that is demanding by 2026 standards, and the 200 MB footprint is consistent with a thin native client that does its heavy lifting in the cloud.

One nuance worth flagging is the distinction between the Mac app and Gemini in Chrome. The release notes describe Gemini in Chrome as a separate rollout currently limited to Google AI Pro and Ultra subscribers in the US, with Chrome's language set to English, and explicitly not available to Workspace business and education plans. The Mac app's eligibility is broader. Workspace administrators evaluating both should not assume a single policy covers them.

The features that did not ship

The most honest part of the launch story is what is missing. Fello AI's review lists three absences that matter: Gemini Live is not in the desktop app, there is no chat folder organization, and there is no multi-account support. The last one is the sharpest. ChatGPT and Claude both handle multi-account switching gracefully, and for anyone juggling a personal Google account alongside a Workspace identity, the lack of in-app switching is a daily friction point rather than a theoretical one.

Gemini Live's absence is harder to read. Live is one of the features Google has used to differentiate Gemini in marketing, and shipping a Mac app without it suggests either an architectural decision to keep voice-first interaction on mobile or a feature gap the team will close in a follow-up release. Google has not committed publicly to a timeline.

Chat folder organization is the smallest of the three gaps but the one most likely to bite power users. Anyone running multiple ongoing projects through Gemini will find the flat history list cumbersome, and it is the kind of feature competitors have offered for over a year.

How this fits into the Apple relationship

The Mac app does not exist in isolation from a much larger story. 9to5Mac reports that Gemini is set to power an upgraded, more personalized version of Siri under a multi-year Apple–Google collaboration announced on January 12, 2026, with further details expected at WWDC 2026 beginning June 8. That makes the timing of a polished native Mac client conspicuous rather than coincidental.

For Apple, Gemini becomes the model behind a Siri overhaul that has been promised and delayed for years. For Google, the Mac app is both a standalone product and a way to establish brand presence on Apple hardware before the Siri integration arrives. A user who has spent two months using Gemini on macOS through the dedicated app is a user who will accept Gemini answers inside Siri as a continuation rather than a surprise.

The sequencing is unusual. Google rarely ships a dedicated Mac app for a flagship service ahead of an OS-level integration with Apple, because doing so risks training users into a habit Apple's own software will eventually try to replace. The fact that Google did it anyway suggests the company sees the Siri integration as additive rather than substitutive, or that the WWDC announcement will reveal a Siri layer that explicitly hands off to the dedicated app for heavier work.

The competitive math for Workspace users

Fello AI's review captures the strategic stake plainly: "If you're a Google Workspace user who's been dabbling with ChatGPT because Gemini was always stuck in a browser tab, this app changes your calculus." That is the audience Google is most directly trying to retain. Workspace customers had a structural reason to keep ChatGPT or Claude open on the desktop, because those apps lived in the Cmd+Tab switcher and Gemini did not.

The Pro tier is priced at launch.99 per month at launch, matching ChatGPT Plus exactly, per Fello AI. That parity is not accidental. It removes price as a decision variable and forces the choice onto feature and integration grounds, which is the comparison Google would rather have.

The counterargument Fello AI raises is fair: "ChatGPT also exists and has computer use" is a real capability gap. OpenAI's agent features let ChatGPT take actions inside applications, not just observe them. Gemini's screen sharing reads the screen and answers questions. That distinction will matter to anyone whose workflow involves delegating multi-step tasks rather than asking for analysis.

What the launch signals

Shipping a native Swift app in under 100 days, by Google's own account, is the kind of timeline that used to be associated with startups rather than platform companies. Whether that pace holds for follow-up releases is the question that will determine whether multi-account support, Gemini Live, and folder organization arrive in months or quarters.

The Mac app's existence is the headline. Its incompleteness is the asterisk. Both will matter as WWDC approaches and the shape of the Apple integration becomes clearer.