What is Google’s Project Mariner?

There are plenty of AI-driven tools available, but most of them still don’t integrate seamlessly into the way we navigate the web or use digital platforms. What we need is an AI that can understand the bigger picture—one that can reason across tasks, analyze data in real-time, and take action without you having to guide it step by step. This is the future of web interaction, where AI adapts to our needs, connects the dots, and works in harmony with the digital world.

Key takeaways:

  • Project Mariner is built on Gemini 2.0 and is designed to automate tasks and enhance user interaction with websites directly in the browser.

  • It processes text, images, and code, enabling users to interact with websites using voice commands or visual and textual input.

  • Mariner navigates websites, performs tasks like filling forms or finding contact information, and provides step-by-step reasoning for its actions, ensuring transparency.

  • Operates only within the active browser tab, requests user confirmation before sensitive actions, and prioritizes security and control.

  • Mariner shows strong results (83.5% success in real-world web tasks) but is still in early development, with improvements planned for accuracy and efficiency.

Project Mariner

Google’s Project Mariner is a research prototype designed to explore the future of human-agent interaction. It is built on DeepMind’s Gemini 2.0 and aims to transform users’ interactions with the web, starting with their browser. This experimental AI tool is embedded as a Chrome extension to perform tasks in your browser.

“The pace of progress in artificial intelligence (I’m not referring to narrow AI) is incredibly fast. Unless you have direct exposure to groups like Deepmind, you have no idea how fast—it is growing at a pace close to exponential. The risk of something seriously dangerous happening is in the five-year time frame. 10 years at most.”— Elon Musk

Project Mariner processes everything on the screen, including pixels and web elements and helps users by handling tasks directly within the browser. It can interpret instructions, reason through complex requests, and take action—all while keeping you in control.

How does Project Mariner work?

Once installed as a Chrome extension, users can communicate with Project Mariner via a chat interface. For example, if you have a list of companies in a Google Sheet, you can ask Mariner to look up contact emails for each one. Mariner will then break down the task into actionable steps, navigate websites to find the required information, and provide you with the results.

A demo presented by Project Mariner shows how the agent displays its step-by-step reasoning as it performs tasks, offering full transparency into the process. Importantly, Mariner only operates within the active browser tab, ensuring that it never runs in the background without your knowledge or control. Before performing sensitive actions, such as placing an order, Mariner seeks user confirmation, giving you peace of mind.

Key features of Project Mariner

With the power of Gemini 2.0, Project Mariner introduces a suite of innovative features that redefine how users interact with their browsers. Below is a list of these key features:

Features of Project Mariner
Features of Project Mariner

1. Native multimodality

  • Project Mariner can understand diverse web elements, including text, images, code, and forms.

  • The AI is capable of processing both visual and textual information, allowing for rich interactions across websites.

  • It can seamlessly execute voice commands, giving users a more natural and hands-free way to interact with their browsers.

2. Browser interaction

  • Mariner autonomously navigates complex websites, automating tasks in real time on behalf of the user.

  • It can perform repetitive actions, saving users valuable time by completing tasks such as finding contact information, filling out forms, or even researching products.

  • When unclear about a task, Mariner will ask for clarification, ensuring accuracy before proceeding.

3. Reasoning and task execution

  • Project Mariner breaks down complex instructions into actionable steps, offering transparency in its decision-making process.

  • It understands the relationships between different web elements (e.g., buttons, forms, links) and ensures tasks are executed correctly.

  • Users are shown a step-by-step view of their reasoning and planned actions, making it easier to understand how tasks are performed.

4. Safety

  • While automating tasks, Mariner operates only within the active browser tab, preventing unwanted actions in the background.

  • Before performing sensitive actions, such as purchasing items, it requests user confirmation, ensuring a layer of safety and control.

  • As part of its development, Google is conducting research to identify potential risks and mitigations, ensuring that the technology is built safely and responsibly with humans in the loop.

Performance and benchmarking

Project Mariner has been evaluated against industry benchmarks to assess its capabilities:

  • WebVoyager benchmark: This test measures the performance of autonomous agents interacting with real-world websites. Project Mariner achieved an 83.5%https://deepmind.google/technologies/project-mariner/ success rate, demonstrating its strong performance as a single-agent setup when it comes to navigating and completing web-based tasks.

  • ScreenSpot benchmark: Mariner scored 84.0% in multimodal screen understanding, proving its ability to analyze graphical user interfaces (GUIs) across different platforms.

While these results are promising, it’s important to note that Project Mariner is still in its early stages. Although the tool demonstrates strong potential, its accuracy and speed are still evolving, with improvements expected over time.

Future implications of Project Mariner

Currently, Project Mariner is available to a small group of trusted testers using the experimental Chrome extension. This limited rollout will help refine the tool’s capabilities and gather feedback to improve its performance.

As Google continues to enhance its agentic AI technologies, Project Mariner represents a major step toward making human-AI collaboration in the browser a reality. Interested users can join the waitlist for the trusted tester program, and in the meantime, Google is also engaging with the broader web ecosystem to ensure that the technology can be integrated smoothly into real-world web environments.

With ongoing improvements and responsible development practices, Project Mariner has the potential to redefine how users interact with the internet. It could make tasks more efficient and give people more time to focus on what matters.

Learn more about generative AI

If you’re fascinated by what generative AI like Project Mariner can achieve and want to dive deeper into the technology, check out our exclusive courses. Gain hands-on expertise and learn how to create your own AI-powered solutions!

Frequently asked questions

Haven’t found what you were looking for? Contact Us


What is Gemini 2.0 Flash experimental?

DeepMind’s Gemini 2.0 Flash that builds on the success of the earlier Gemini 1.5 Flash. It improves performance, offering faster response times and better benchmark results. The model supports multimodal inputs (text, images, video, and audio) and outputs (text combined with images and steerable text-to-speech).


What is the Google Glass project?

Google Glass is a wearable augmented reality (AR) device that displays information directly onto a small, transparent screen in front of the user’s eye.


What is Google’s Project Astra?

Project Astra is a Google research prototype aimed at creating a universal AI assistant. It integrates AI into devices like smartphones and prototype glasses to help users explore and interact with their world in new ways. Currently, it’s available for limited testing, with users able to join a trusted tester waitlist.


Free Resources

Copyright ©2025 Educative, Inc. All rights reserved