Project MarinerGoogle's experimental AI agent that browses and acts on the web for you.

4.2 (5)

Reviewed by Daniel Nikulshyn·Updated July 2026

AI Agent Web Automation Browser Extension Research Preview Google DeepMind Gemini Productivity

1 / 4

Overview

Project Mariner is an experimental web-browsing AI agent from Google DeepMind that can understand on-screen content and take actions inside a browser to complete tasks on a user's behalf. Built on Gemini, it interprets natural language goals and translates them into sequences of clicks, scrolls, form fills, and navigation steps across live websites. The agent is designed for multi-step workflows such as researching products, gathering information, filling out forms, or assembling shopping carts. It runs in a Chrome extension and keeps the user in the loop, surfacing its reasoning and asking for confirmation before sensitive actions like purchases. As a research prototype, Mariner is being tested with a limited group of trusted users while Google studies safety, reliability, and the broader implications of agents that operate autonomously on the open web.

Key features

Autonomous web navigation and clicking
Natural language task instructions
Chrome extension interface
Step-by-step action transparency
Confirmation prompts before key actions
Built on Google's Gemini model

Pricing

Model: Freemium
Category: AI Agents
Rating: 4.2 / 5 (5)

Use cases

Automate Product Research

Delegate multi-site product comparisons to Mariner, which browses listings, gathers specs and prices, and summarizes findings so you can decide faster.

Assemble Shopping Carts

Give Mariner a shopping list in natural language and let it navigate retailer sites, add items to the cart, and pause for your confirmation before checkout.

Fill Out Repetitive Web Forms

Hand off tedious form-filling tasks across multiple pages, with the agent clicking, typing, and navigating while you supervise its steps in Chrome.

Gather Information Across Sites

Ask Mariner to collect data from several websites for a research goal, watching its reasoning and actions unfold step by step in the browser.

Pros & Cons

Pros

Automates tedious multi-step browser tasks
Powered by Gemini's reasoning capabilities
Keeps humans in the loop for sensitive actions
Operates on real, live websites

Cons

Limited research preview with restricted access
Can be slow on complex pages
May struggle with dynamic or unusual site layouts
Requires user supervision for reliability

Reviews

4.2

Average from 5 ratings.

Liam O’Connor

Apr 13, 2026

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on autonomous web navigation and clicking, and automates tedious multi-step browser tasks caught me off guard. Can be slow on complex pages is why this isn't a perfect score, still, I'd recommend giving it a real trial.

Naomi Suzuki

Sep 28, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is step-by-step action transparency — handled better than most — and powered by Gemini's reasoning capabilities. Worth the time if this is your use case.

Margaret Whitfield

Jul 29, 2025

Compared a few options

Evaluated this against two competitors. Where it wins: built on Google's Gemini model and powered by Gemini's reasoning capabilities. Where it lags: may struggle with dynamic or unusual site layouts. On balance the feature set — especially step-by-step action transparency — justifies the 4 stars for our use case.

Omar Haddad

Jul 14, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is natural language task instructions — handled better than most — and automates tedious multi-step browser tasks. Requires user supervision for reliability is my one real gripe. Worth the time if this is your use case.

Kwame Mensah

Jul 13, 2025

Skeptical, then convinced

I went in skeptical — most tools in this space overpromise. It actually delivers on chrome extension interface, and powered by Gemini's reasoning capabilities caught me off guard. May struggle with dynamic or unusual site layouts is why this isn't a perfect score, still, I'd recommend giving it a real trial.

Q&A

Can Mariner handle complex or dynamic website layouts?

Mariner may be slower on complex pages and can struggle with highly dynamic or unusual site layouts, so it works best on standard, well‑structured web pages and still requires user supervision for reliability.

Asked by Hasan Demir · Aug 24, 2025

How does Mariner handle sensitive actions like purchases?

The agent surfaces its reasoning step‑by‑step and asks for your explicit confirmation before performing any key actions such as purchases, keeping you in the loop for safety and reliability.

Asked by Zeynep Aydin · Aug 12, 2025

What browsers does Mariner work with, and how is it installed?

Mariner runs as a Chrome extension, so you need the Google Chrome browser. Install the extension from the provided preview link, and it will appear in your browser toolbar for use.

Asked by Marisol Pena · May 26, 2025

How can I get access to Project Mariner and is there a cost?

Project Mariner is currently a research prototype available only to a limited group of trusted users. Google has not published public pricing, and access is granted by invitation during the preview phase.

Asked by Jasper Vermeer · May 5, 2025