Kernel Sphere – A web agent that reads, reasons, and acts in a real browser

Summary: Kernel Sphere is an open-source DOM-native automation platform that uses Google Gemini and Playwright to execute tasks described in plain English by navigating real browsers and returning structured results. It handles complex web interactions but currently achieves 48% accuracy on diverse tasks, highlighting areas for improvement.

What it does

It opens a real browser at a given URL, reads the page content, decides actions step-by-step, and outputs structured data based on the described task.

Who it's for

Developers needing automated web interactions that involve reasoning and dynamic page handling, especially those interested in contributing to open-source AI agent development.

Why it matters

It addresses challenges in automating complex web workflows like login flows, CAPTCHAs, and multi-step forms that traditional agents struggle with.