โ€” Autonomous Agent

Google Project Mariner

Last updated July 21, 2026 ยท Reviewed by ToolForge Editorial

Google's Chrome-embedded AI agent โ€” Gemini 2.0 lives inside your browser tab and drives the page for you. The direct answer to OpenAI Operator.

โ˜… 3.8/5 ยท Limited preview ยท Since December 2024 ยท Waitlist

Google's bet: the agent lives in your browser, not the cloud

Project Mariner was announced in December 2024 alongside Gemini 2.0 and has been slowly rolling out to waitlist users throughout 2025-2026. The core idea is different from Operator's: instead of running in OpenAI's cloud and screen-sharing a remote browser, Mariner is a Chrome extension that drives your actual browser โ€” your logged-in tabs, your saved passwords, your cookies. This is more powerful (it can do things Operator can't, like accessing your bank session) and more dangerous (so can a malicious agent). Google has been cautious with rollout for exactly this reason.

Who it's for: Power users who got off the waitlist. There's no paid tier yet โ€” Google is deliberately throttling access while it figures out the safety story. If you're in, you're effectively a beta tester providing Google with usage data.

Key features

Tab Runs in your real Chrome

Mariner is a Chrome extension, not a remote browser. It can interact with your logged-in sessions โ€” Gmail, your bank, internal corporate apps โ€” without you re-authenticating. Operator literally cannot do this.

Multi-tab Up to 10 tabs at once

Mariner can drive up to 10 Chrome tabs simultaneously โ€” researching across sites, comparing options, and copying data between them. Operator is single-threaded.

Gemini 2.0 Multimodal reasoning

Built on Gemini 2.0, which means native multimodal reasoning โ€” it can read images, charts, and PDFs inline on a page, not just text. Strong on tasks like "summarize the financial report in tab 3 and add the key numbers to my spreadsheet in tab 5."

Confirm Sensitive-action prompts

Like Operator, Mariner pauses for confirmation on payments, send-email actions, and any URL on Google's blocklist (banks, government sites, etc.). You see the proposed action before it executes.

Workspace Google account integration

Because it lives in your browser, Mariner has direct access to your Google account โ€” Docs, Sheets, Drive, Calendar. Output destinations are first-class: "put the results in a new Google Sheet" just works.

The honest take

โœ“ What works

  • Drives your real logged-in browser โ€” no re-auth, works on private/internal apps
  • Multi-tab parallelism genuinely faster than Operator's single-threaded model
  • Native Google Workspace integration is excellent if you live in Docs/Sheets
  • Free during preview โ€” no $200/mo Pro tier gate
  • Gemini 2.0 multimodal is strong on pages with charts/images, not just text

โœ— What doesn't

  • Waitlist is brutal โ€” most users still don't have access in mid-2026
  • Chrome-only โ€” no Firefox, Safari, or Edge support
  • Security model is unresolved โ€” Google throttles features precisely because driving your real session is risky
  • Slower than Operator on simple tasks (more safety checks per action)
  • No clear pricing path post-preview โ€” could become a paid Gemini tier

Verdict

Mariner is technically the more capable agent than Operator โ€” running in your real browser with multi-tab parallelism is a meaningful architectural advantage. But Google's caution means almost no one can actually use it in mid-2026. If you're on the waitlist, you're effectively testing a product Google hasn't decided how to ship. For now, Operator wins on availability; Mariner wins on capability, when you can get it.

๐Ÿ’ก Transparency: This review contains affiliate links. If you sign up through our link, we may earn a commission at no cost to you. We only recommend tools we use ourselves. Full disclosure.

Related Tools

Try Project Mariner

Free in preview ยท Join the waitlist ยท Chrome desktop only

Join waitlist โ†’