Project Mariner

Project Mariner has emerged as a notable solution in the AI tools landscape, offering project mariner is an experimental ai browser agent developed by google deepmind that autonomously interacts with web pages, performs multi‑step tasks, and navigates online workflows on behalf of the user based on natural language instructions.. With a free tier available, it's accessible to teams of all sizes.

Project Mariner is an agentic AI prototype designed to observe, plan, and act within web browsers to complete tasks that would normally require human interaction. Built around Google’s advanced Gemini foundation models, it captures the current state of the browser including text, images, code, and form fields and uses natural language understanding and multimodal reasoning to devise an actionable plan. Once a plan is formulated, the agent controls the browser by moving the cursor, clicking buttons, filling forms, and traversing sites to complete complex, multi‑step workflows with minimal input from the user. Mariner also incorporates a “teach and repeat” capability, where it learns workflows demonstrated once and can replicate similar tasks in the future, enhancing productivity for repetitive tasks. In recent updates, it has been adapted to run on cloud‑based virtual machines, allowing it to handle multiple tasks concurrently (up to 10 at a time) while users continue other work. As of mid‑2025, Project Mariner is available to subscribers of Google’s AI Ultra plan in the U.S., and its capabilities are being integrated into the Gemini API and Vertex AI so developers can build applications powered by this agent. Unlike typical browser automation scripts, Project Mariner reasons about web context and user intent to execute tasks that extend beyond simple button clicking.