Project Mariner

A research prototype exploring the future of human-agent interaction, starting with browsers

Automate multiple tasks, simultaneously

Use natural language to assign AI agents to handle time-consuming tasks, like research, planning, and data entry. They can tackle tasks simultaneously in browsers running on virtual machines.

Multimodal reasoning

Project Mariner observes what’s displayed in the browser. It then reasons to interpret your goals, makes a plan — and takes action.

Observes

Identifies and understands web elements including text, code, images and forms, to build an understanding of what is displayed in the browser.

Plans

Interprets complex goals and reasons to plan out actionable steps. The agent will also share a clear outline of its decision-making process.

Acts

Navigates and interacts with websites to carry out the plan, while keeping you informed. You can further prompt the agent at any time, or stop the agent entirely, and take over what it was doing.

Teach and repeat

Once agents have learned how to do a task, they can try to replicate the same workflow in the future with minimal input — freeing up even more of your time.

Coming to the Gemini API

We’re bringing Project Mariner’s computer use capabilities into the Gemini API, and we’re bringing more capabilities to other Google products soon.

Learn more

Building responsibly in the agentic era

We recognize the responsibility it entails to develop these new technologies, and aim to prioritize safety and security in all our efforts.

Learn more

Experience Project Mariner

Project Mariner is now available in the US to Google AI Ultra subscribers. It’s still a research prototype, and we appreciate and encourage feedback as we further develop its capabilities.