Google Launches the Gemini App Globally on Windows 10 and 11

What Google released on September 10

Google announced the Gemini app for Windows on September 10, 2026. The company says the app is available now, globally, on Windows 10 and Windows 11 through gemini.google/desktop. The desktop client is designed to sit alongside other applications rather than requiring a browser tab.

The central interaction is Alt + Space. Google says that shortcut opens Gemini over the active application for tasks such as checking a fact in a document or generating title ideas for a presentation. The announcement also describes a dedicated workspace for longer tasks. These are vendor-described capabilities; AccessAllGPT did not install or operate the app.

The workspace brings existing Gemini surfaces to the desktop

Inside the dedicated app, Google says users can access what they already use Gemini for. The examples include asking Gemini to create a project summary from Gmail and Google Drive, handing a multi-step task to Gemini Spark, creating images with Nano Banana and creating video with Gemini Omni.

That wording matters: this is a new Windows access surface, not evidence of a new foundation model. The launch post does not identify the default model, expose a desktop model selector contract, promise feature parity across accounts or publish an API for controlling the Windows client. Buyers should inventory what appears for their exact account, region and subscription instead of treating the marketing list as a universal entitlement.

There is no new model ID, API contract or desktop price

Google’s launch post contains no model or API identifier, context-window specification, token price, rate limit, latency target, offline mode or enterprise service-level commitment for the Windows app. Developers looking for a callable model have not received a new API surface in this announcement. The relevant unit is an end-user desktop product connected to the existing Gemini service.

The post also does not state a separate price for downloading or using the Windows app. A footnote attached to the Spark and Omni references says a Google AI subscription is required, availability varies and the features are limited to users 18 or older. That condition should not be widened into a claim that every part of the desktop app requires the same paid plan. Verify plan entitlements in the target account before procurement or rollout.

Global availability still has account-level conditions

“Available globally” describes the Windows release footprint in Google’s announcement, but it does not guarantee that every illustrated feature is enabled in every country, language, managed tenant or age group. Google explicitly qualifies Spark and Omni availability. Its Privacy Hub also notes that work and school Google Accounts may be subject to different data-handling terms.

For an organization, test at least one representative consumer account and one managed account only when policy permits. Record operating-system build, account type, region, language, plan and enabled Connected Apps. If a required feature is absent, treat that as the observed entitlement rather than attempting to infer a rollout date from the launch headline.

Connected Apps make permissions part of the product decision

A desktop shortcut is convenient precisely because it reduces the distance between an active task and an AI request. The same convenience can make it easy to send document text, files or connected account data without pausing to inspect the boundary. Google’s example of summarizing information from Gmail and Drive depends on Connected Apps access, not merely a local desktop overlay.

Approve only the connections needed for a defined workflow. Test whether administrators can restrict them, whether users understand which account is active and whether disconnecting an app produces the expected result. Google’s Privacy Hub says Connected Apps and other services may retain information under their own policies, and deleting Gemini Apps activity does not necessarily delete data another service received.

Spark adds a remote-computer data boundary

Google describes Gemini Spark as a personal AI agent for multi-step tasks. The Privacy Hub says Spark can save and execute Markdown files, code and other task information, including data from Personal Intelligence and Connected Apps. It also says Gemini may share information necessary to complete tasks with other services and third parties through Connected Apps and the remote browser.

Teams should therefore evaluate Spark separately from ordinary question answering. Start with non-sensitive, reversible work; review which files and accounts it can reach; inspect the remote browser and remote computer controls; and require a human checkpoint before external communication, purchase, deletion or publication. Google warns that task features can make mistakes, including purchases or unexpected sharing. That warning is a reason to narrow authority, not evidence that a particular failure will occur.

Activity settings change—but do not erase—the retention boundary

Google says that with Keep Activity on, chats and shared material can be saved in Gemini Apps Activity and used to provide, develop and improve services, including generative AI models, with human reviewers involved in some review. The default auto-delete period described by the Privacy Hub is 18 months, with options including 3 or 36 months or no automatic deletion.

When Keep Activity is off, Google says future chats do not appear in activity and are not used to train its AI models unless feedback is submitted, but future chats are still saved for 72 hours so the service can respond, process feedback and protect users and the public. Reviewed material can have separate retention. Before workplace use, classify allowed data, configure the account, validate managed-account terms and give users a simple rule for content that must never be submitted.

What to test before a managed rollout

First verify installer provenance and signature from Google’s canonical desktop link, then install on supported Windows 10 and Windows 11 builds in a disposable test profile. Confirm update behavior, startup settings, resource use, proxy compatibility, endpoint-management visibility, uninstall behavior and whether Alt + Space conflicts with PowerToys, language tools, launchers or accessibility software.

Next freeze a small task set: overlay questions about an open document, a Gmail-and-Drive summary with planted irrelevant records, image generation, a refused or unavailable feature, a disconnected account and a Spark task with an explicit stop point. Record what data is transmitted, which permissions are requested, what appears in activity, whether deletion behaves as documented and whether the app preserves the correct Google account. AccessAllGPT did not run these tests; they are a practical acceptance plan.

The decision: allow, pilot, constrain or wait

Allow the standard app for a defined user group when installer provenance, account controls, permitted data and required features are verified. Pilot Connected Apps and Spark separately because they widen the data and action boundary. Constrain the app to non-sensitive drafting and research when managed controls or deletion evidence remain incomplete. Wait when the canonical installer cannot be verified, required enterprise terms are unclear or a core feature is not enabled for the target account.

The immediate practical change is access: Windows users now have a first-party Gemini client and a system-wide shortcut. The announcement does not by itself change a model-selection decision or establish that the app is faster, lighter or safer than the web client. Treat those as testable deployment questions, and revisit the record when Google names the underlying model contract, changes entitlements or adds the promised native desktop capabilities.

Disclosures

AccessAllGPT did not receive early access, a Google AI subscription, installer, device, briefing, test data or compensation for this article. Google did not sponsor, review or endorse it. AccessAllGPT Research is operated by NeuralArc, is independent, and is not affiliated with Google, Google DeepMind, OpenAI or organizations cited. Publication-wide relationships are listed on the disclosures page.