Google brought Gemini Spark, its agentic assistant, to the Gemini app for macOS on July 1, extending the tool from mobile and web into a desktop agent that can act on files stored locally on a user's Mac.

What's new on the desktop

On macOS, Spark can be asked to sort large batches of PDFs sitting in a Downloads folder into specific subfolders, or to pull numbers from invoices saved on the computer into a new budget spreadsheet in Google Workspace, bridging local files and cloud apps in a single request. Google also rolled out custom Model Context Protocol (MCP) support, letting users connect additional third-party apps into Spark beyond the built-in integrations. At launch, Spark connects to Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals, enabling tasks like designing a flyer, reordering groceries, or booking a table without leaving the assistant. Google also gave Spark the ability to track evolving topics in real time, such as sports scores, market moves, or breaking news, and surface updates as they happen. The company says a future update will let users assign multi-step tasks to Spark from their phone that pull from files on their Mac, such as fetching a sales report and emailing a revenue figure, though that remote-trigger capability is not yet available.

Availability

Gemini Spark for macOS is in beta and limited to Google AI Ultra subscribers, Google's $250-per-month tier, who are 18 or older and located in the US.

Why it matters

The macOS launch pushes Spark into more direct competition with other desktop AI agents, including Claude Desktop and Microsoft's Copilot, all of which are racing to move beyond browser-based chat into agents that can read and act on a user's local files and installed apps. Restricting the desktop beta to Ultra subscribers signals Google is treating deep file-system and app access as a premium, trust-sensitive capability rather than a mass rollout, at least for now.