Gemini Spark Beta Adds Persistent Mac Automation for $99 a Month
Google's agentic assistant expands to macOS Sequoia with local file system integration, Model Context Protocol support, and cloud-backed background execution.
Google officially launched Gemini Spark for the Gemini macOS app, moving its proactive AI assistant onto desktop hardware. Starting with Gemini app version 1.80.15.516, the agentic system can execute multi-step workflows that bridge the local file system and Google Workspace.
Desktop Automation Architecture
Spark operates on Google’s Antigravity platform using Gemini 3.5 Flash. Rather than executing completely on the device edge, the agent relies on dedicated cloud virtual machines. This architectural split allows tasks to persist in the background. A user can initiate a heavy file-sorting operation and close their laptop, and the remote VM will orchestrate the job until completion.
Deep integration with macOS requires explicit permission bounding. Users must assign specific directories to the agent via a dedicated Spark tab in the sidebar. The agent cannot traverse unlinked parts of the local drive. Once linked, the agent handles batch renaming, context-based sorting, and cross-application data extraction. For example, it can parse figures from local PDF invoices and populate a remote Google Sheet autonomously.
Spark also monitors live topics in the background. The agent tracks custom queries around finance, weather, and sports, pushing alerts to the user when predefined conditions trigger. Google also outlined an upcoming remote execution pipeline, which will allow users to initiate local Mac file retrievals directly from the Gemini mobile app.
Extensibility and Integrations
To scale functionality beyond Workspace, Google is rolling out support for custom Model Context Protocol (MCP) servers. This gives developers a standardized path to connect internal tools and custom databases to Spark’s reasoning engine directly on the desktop.
The update ships with native connectors for Google Tasks, Google Keep, Canva, Dropbox, Instacart, OpenTable, and Zillow Rentals. Integrating Tasks and Keep resolves a notable constraint from the initial Spark preview, allowing the agent to manage personal to-do lists and notes natively.
System Requirements and Pricing
Access is restricted to early adopters in the United States on specific hardware and billing tiers.
| Requirement | Specification |
|---|---|
| Operating System | macOS Sequoia (15.0) or later |
| Hardware | Apple Silicon Mac |
| App Version | Gemini desktop 1.80.15.516+ |
| Base Pricing | $99 per month (Google AI Ultra) |
| High-Volume Tier | $199.99 per month |
The $99 baseline tier includes 20TB of cloud storage and standard access to the Antigravity agentic platform. The $199.99 tier increases storage to 30TB and expands usage limits for high-volume agentic tasks.
Granting a persistent, cloud-backed agent read and write access to local file structures introduces distinct security tradeoffs. If you plan to deploy custom MCP integrations for your desktop environment, audit the data exposed within your linked Spark directories to ensure sensitive materials remain isolated from the agent’s view.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
How to Automate Agent Evaluation With Google Quality Flywheel
Learn how to configure Google's new Agent Quality Flywheel skill to automate evaluation, grading, and prompt optimization for your AI coding agents.
Google Launches Workspace Intelligence and Workspace MCP Server
Google announced a new agentic AI layer for Workspace apps alongside hardware updates and a Model Context Protocol server for developers.
IBM Pivots to Agent Logic to Control Multi-Step AI Workflows
A joint technical publication from IBM and Hugging Face details how strict state management and formal logic layers can govern long-running enterprise agents.
Frontier Agents Score Below 50% on SRE Task Benchmark
IBM Research and Artificial Analysis launched ITBench-AA, revealing that top frontier AI models score below 50% on complex enterprise SRE tasks.
Android XR Launches With Gemini 3.5 Wearable Agent Support
Google's Android XR platform introduces a two-tier hardware strategy for smart glasses, relying on Gemini 3.5 to process multimodal agentic workflows.