browser-agent

Why we included this project

Magnitude automates the browser by looking at the actual screen: it hands the live screenshot to a vision model that plans actions in pixel coordinates, an approach that tends to handle modern, complex interfaces more reliably than scripts tied to page structure. A single natural-language command can drag cards across a project board, and the same engine pulls out structured records that match a zod schema, covering both high-level goals and precise low-level actions. It also doubles as a test runner with visual assertions, so it fits end-to-end web testing as well as one-off automation like moving data between apps that have no API. Since it is built on Playwright, developers who want finer control can reach into the underlying page and context objects directly.