Gemini Spark is Google's brand-new autonomous AI agent, announced at Google I/O 2026 — and this is the desktop client that takes it beyond Gmail and Calendar into full system-wide control of your Windows or Mac. All the power of Spark — 24/7 background workflows, persistent memory, custom Skills, proactive task automation across Connected Apps — plus the one thing the official Gemini app can't do: act on every application, file, and window on your computer. Open the desktop app. Say "clear my newsletter clutter, prep my Monday brief, then refactor the marketing folder by client." Spark Desktop handles all of it in the background while you sleep. No Gateway setup. No ngrok tunnel. No messaging-app config. Just install, sign in once, and your agent is alive. One-click signed installer, free launch tier via Google partnership, Local Library mode for privacy-first users.
Install · Features · How it works · Comparison · Privacy · FAQ · Roadmap
Google launched Gemini Spark at Google I/O 2026 — a persistent, always-on AI agent built on Gemini 3.5 and Google's Antigravity agent infrastructure. Spark is Google's answer to the autonomous-agent paradigm that OpenClaw turned into a 250,000-star phenomenon over the past four months. Custom Skills, proactive automation, recurring workflows, Connected Apps integration, Personal Intelligence memory — Spark brings that full pattern to two billion Gemini users with zero setup.
But the official Spark, by design, lives inside Google's ecosystem: Gmail, Calendar, Drive, Docs, Meet, Chrome, Workspace. That's a powerful sandbox — but it's still a sandbox. The whole point of an autonomous agent is that it can act on your workflow, which means every app, every file, every window on your machine. Local PDFs you haven't uploaded to Drive. Code in your IDE. Design files in Figma desktop. Recordings in your screen-capture tool. Spreadsheets in Excel, not Sheets. The agent that can clean up your inbox should also be able to organize your local Downloads folder.
This desktop client takes Spark's brain and gives it the rest of your computer. Through our Brand Approved partnership with Google, every Spark capability — Skills builder, scheduled workflows, Connected Apps, Personal Intelligence — runs natively on Windows and Mac, plus we add system-wide control: any app, any file, any window. Type your goal in plain English, watch the agent execute across your entire desktop. OpenClaw-grade autonomy with Spark-grade reliability and zero setup pain.
What you get:
- Full Gemini Spark feature set — Skills, scheduled workflows, Connected Apps, Personal Intelligence memory, 24/7 background execution
- System-wide control — Spark agent acts on every app, file, and window on your Mac or PC, not just Google Workspace
- Zero setup — no Gateway, no ngrok, no messaging-app config (unlike OpenClaw and similar open-source agents)
- Native desktop app for Windows and Mac, one-click signed installer
- Free launch tier — generous daily allowance through our Google partnership, no credit card required
- Local Library mode — Skills, history, and memory stored on your machine, never synced
- MIT licensed, signed installers, fully auditable source, zero telemetry
Windows: Download Gemini-Spark-Setup_x64.7z from the latest release and double-click. Digitally signed, passes SmartScreen.
Mac: Download Gemini-Spark_macOS.dmg, drag to Applications. Apple Developer ID signed and notarized. Universal binary (Apple Silicon M1–M5, Intel).
90-second flow: Open the app. Sign in with your Google account (Connected Apps auto-discover Gmail, Calendar, Drive, Docs, Meet). Grant system permissions (accessibility on Mac, UI Automation on Windows) so Spark can act beyond the browser. Create your first Skill in the Skill Builder, or pick from 50+ starter Skills in the library. Hit Activate — Spark is now your 24/7 agent.
- Custom Skills — define repeatable workflows in plain English; Spark turns them into reusable agent templates
- Scheduled workflows — daily, hourly, on-trigger; Spark runs autonomously without prompting
- Connected Apps — native integration with Gmail, Calendar, Drive, Docs, Meet, Tasks, Keep, Chrome through Google's official API
- Personal Intelligence — Spark builds a context layer from your usage patterns, learns your preferences, gets smarter over time
- Multi-step task execution — break a goal into substeps, execute across apps, surface only what needs your decision
- Proactive automation — Spark notices recurring patterns and offers to automate them
- Background heartbeat — checks tasks at configurable intervals (default every 5 minutes), acts or waits
- Full desktop control — open any application, click any button, type into any field, read any window
- Local file operations — search, organize, rename, move, summarize files anywhere on your disk
- Cross-app workflows — read from Excel, write to Notion, screenshot from Figma, attach to email — all in one Skill
- Window-aware reasoning — agent sees what you see; understands the context of the active app
- Shell command execution (with permission) — run terminal commands for power-user automations
- Code-aware actions — read your codebase, run tests, commit changes, open PRs through Git CLI integration
- OS event triggers — Skills can fire on file changes, USB events, network state, clipboard updates
50+ starter Skills included at launch, covering the most-requested OpenClaw and Spark use cases:
- Inbox triage — daily Gmail decluttering, newsletter archiving, spam filtering
- Meeting prep — pull calendar, fetch related docs, brief you 15 min before each meeting
- Daily digest — personalized news, GitHub notifications, Slack summaries
- Code review nightly — scan your repo for outdated dependencies, security issues, broken tests
- Local file janitor — clean Downloads, organize Desktop, archive old files by rule
- Receipt collector — find receipts in email, save to Drive, categorize for tax season
- Research watcher — track topics across web/arXiv, summarize weekly
- Travel prep — flight reminders, hotel check-ins, itinerary synthesis
- Social-media scheduler — draft posts from drafts folder, queue across platforms Build your own in the visual Skill Builder, or write them in plain English in the conversation tab — Spark turns natural language into agent workflows automatically.
Spark Desktop runs as a lightweight system tray / menu bar app. The heavy reasoning happens on Google's cloud through the Spark API; the local agent handles execution, system control, and local context. You can close every window — Spark keeps working. Wake up to a clean inbox, a Monday brief, and a Slack message saying "I noticed three meetings tomorrow conflict with your gym block, want me to reschedule the gym?"
Spark Desktop integrates with Google's Gemini Spark through the Brand Approved partnership API — every agent action routes through the official Spark backend running on Gemini 3.5, so you're using the real model and real Skills infrastructure, not a wrapper. The desktop layer handles three things Google's browser Spark can't:
- System-wide execution — operating system APIs (Windows UI Automation, Mac Accessibility) let the agent see and act on any application, not just Chrome tabs
- Local context — file system access, clipboard monitoring, screen state awareness all happen on your machine
- Background reliability — runs as a system service, survives reboots, doesn't depend on a browser tab being open Two-tier model routing automatically picks the right model for each step:
- Gemini 3.5 Flash for fast classifier steps, simple replies, file operations
- Gemini 3.5 Pro for multi-step reasoning, complex Skill execution, decision points You can override per-Skill if you want manual control over cost vs. quality.
| Feature | Gemini Spark (this app) | Gemini Spark (Google web) | OpenClaw | Anthropic Cowork | OpenAI Operator |
|---|---|---|---|---|---|
| System-wide control (any app) | Yes | Browser & Workspace only | Yes (with setup) | Limited | Yes |
| Zero setup required | Yes | Yes | No (Gateway, ngrok, config) | Partial | Partial |
| Native desktop app | Yes | No (browser) | Self-hosted | Yes | No (browser) |
| 24/7 background execution | Yes | Yes | Yes | Yes | Yes |
| Connected Apps (Gmail, Calendar, Drive) | Native | Native | Via skills | Via tools | Limited |
| Skill Builder | Yes (visual + plain English) | Yes | Manual code | Code-based | Limited |
| Local Library mode | Yes | No | Yes (local-first) | No | No |
| Free launch tier | Yes | Gemini Advanced tier | BYO API key | Paid | Paid |
Spark by default uses your Google account memory and Connected Apps — that's how it gets smart fast. But if you're working on NDA projects, sensitive client work, or simply prefer local-first, you have a choice.
- Cloud Library mode (default) — Skills, history, and Personal Intelligence stored in your Gemini account, synced across devices
- Local Library mode — Skills, history, and memory stay on your machine in encrypted local storage. The agent still calls the Spark API for reasoning, but your library, patterns, and context never sync anywhere
- Granular permissions — control which apps, files, and folders Spark can access at any time
- Action audit log — every action Spark takes locally is logged and reviewable; one-click rollback for file operations
- Sensitive action prompts — purchases, payments, irreversible deletes, external messages all require explicit confirmation by default (unlike the official Spark which warns it "may share your info or make purchases without asking")
- No app-side telemetry — zero analytics, zero tracking
- Signed installers — code-signed Windows .exe, Apple Developer ID + notarization on Mac, SHA-256 checksums published
- Open source — MIT licensed, every line auditable
Is it really free? What's the catch and what happens after launch tier? The app is 100% free, MIT licensed, no subscription, no premium tier, no telemetry. Agent actions through our launch partnership tier with Google are free during the launch window — generous daily allowance, no credit card required. After the launch tier closes, you connect your own Google AI Studio or Gemini Advanced subscription. Heavy users (running constant background workflows) typically pay Google a few dollars per month at standard Spark rates. The app itself stays free forever.
Is this an official Google app? Is it safe to download? This is a third-party desktop client built specifically for Gemini Spark through our Brand Approved partnership with Google. It uses the official Spark API, so every agent action runs on the real model. The installer is code-signed on Windows and notarized with an Apple Developer ID on Mac, with SHA-256 checksums published. The full source is MIT licensed on GitHub for security audit. As a general 2026 rule, autonomous-agent installers carry real risk — CrowdStrike has published warnings about misconfigured OpenClaw deployments being commandeered as backdoors — so verify signatures and download only from the Releases page of this repo.
How does this compare to OpenClaw, Cowork, and OpenAI Operator? Different strengths. OpenClaw remains the most powerful self-hosted option — community-extensible skills, fully local, model-agnostic — but it requires Gateway setup, ngrok tunnels, and messaging-app configuration, which limits adoption to developers. Cowork (Anthropic) excels at desktop tasks within Claude's ecosystem. OpenAI Operator handles browser-based agentic workflows well. This desktop client wins on three axes: zero-setup full-desktop autonomy (no Gateway, no config), native Google Workspace depth through Spark's official integrations, and OpenClaw-style system-wide control as a single combination — which none of the others currently offer.
What happens to my data and my files? Can the agent make purchases without asking? The official Spark warns it "may share your info or make purchases without asking" — this desktop client adds an explicit safety layer on top. By default, purchases, payments, irreversible deletes, external messages, and credential-sharing actions all require confirmation, even when the underlying Spark API would have done them autonomously. You can disable confirmation per Skill if you want full automation, but the default is paranoid. Combined with Local Library mode and the per-folder permission system, this is the rare AI agent setup that fits NDA and enterprise workflows without legal review.
What can Spark Desktop actually do that the browser Gemini app can't?
Three big buckets. First, every application on your computer, not just Chrome — Spark Desktop can read Excel, control Figma, edit Notion, run terminal commands, manage local files in any folder. Second, background system services — Skills fire on file changes, USB events, clipboard updates, network state shifts, not just on schedule or user prompt. Third, full local context — your Downloads folder, your ~/Documents, your IDE workspace, your screen recordings, all feed into Spark's reasoning when you want them to. The browser Spark sees Workspace and the active tab; Spark Desktop sees your machine.
v1.1 — Linux packages (.deb, .rpm, AppImage). Community Skill Marketplace (share and import Skills as .skill bundles). Voice activation through OS-native dictation.
v1.2 — Multi-machine sync (your agent state across desktop and laptop). Team Skills (shared workspaces for org-level automation). Plug-in SDK for third-party Connected Apps beyond Google Workspace.
v2.0 — Self-hosted enterprise edition with on-prem deployment. Granular role-based permissions for managed teams.
MIT License. See LICENSE.
This is a third-party desktop client built for Gemini Spark, Google's autonomous AI agent announced at Google I/O 2026. Brand assets and the "Powered by Gemini Spark" badge are used in accordance with Google's Brand Guidelines and our Brand Approval. "Gemini," "Gemini Spark," "Antigravity," "OpenClaw," "Cowork," "Operator," and other product names are trademarks of their respective owners and used here solely to identify the technologies this app integrates with or compares to (nominative fair use). When using this app, agent actions are processed through the official Gemini Spark API under Google's standard privacy and usage policies — this client adds no intermediary server and collects no additional data beyond what your local Skills require. Autonomous agents carry inherent risks: they can make purchases, send messages, modify files, and access connected accounts. Review your Skills and permission settings carefully, and supervise the agent during the first weeks of use until you trust its behavior on your specific workflows.
If Spark Desktop saved you a setup weekend, an inbox afternoon, or a manual workflow you used to dread, please star the repo on GitHub. It's the only metric we track.