Google Launches Gemini Spark For macOS With App Connections

Gemini Spark's jump to the Mac desktop is the clearest signal yet that Google intends agentic assistants to compete for a permanent, file-level presence on a user's machine rather than staying confined to a chat window or mobile app. Anthropic's Claude Desktop and Microsoft's Copilot already occupy this ground on Windows-heavy enterprise fleets, so a native, permissioned macOS agent from Google closes a real gap for teams running Apple hardware, and it puts three major labs on a collision course over which assistant becomes the default automation layer for everyday desktop work.

Google said on June 30, 2026 that Gemini Spark is rolling out in beta to the Gemini app on macOS, limited for now to Google AI Ultra subscribers aged 18 and older in the US. Inside the app, Spark can act directly on desktop files and Google Workspace data: Google's own example has it sorting PDFs from a Downloads folder into designated subfolders, or building a budget spreadsheet from locally saved invoices and then refreshing it on a recurring schedule. Google says Spark only touches files a user explicitly grants it access to. A remote-execution mode is coming soon that would let a user hand Spark a multi-step task from a phone, for instance locate a sales report on the Mac, pull a revenue figure, and email it, and have the assistant complete it unattended.

Alongside the desktop launch, Google is broadening Spark's reach into outside services. New integrations with Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals join existing hooks into Google Tasks and Google Keep, and Google is adding support for custom Model Context Protocol connections, letting users or developers wire Spark into additional tools beyond Google's own catalog. Spark also gains the ability to monitor topics continuously, covering sports scores, stock-price thresholds, and breaking news, and proactively alert a user rather than waiting for a prompt, a shift from reactive chatbot toward standing background agent.

The rollout lands in the same week Meta acknowledged its own agent programs have lagged expectations, underscoring how uneven progress on agentic AI has been across the industry even as vendors race to ship desktop-native versions. For practitioners evaluating assistant platforms, the MCP support is the detail worth watching: it determines whether Spark becomes an extensible automation hub or stays a closed Google product.