OpenAX

The web was built for human eyes, we’re fixing that for agents. Agents use the web by looking at it. They screenshot the page, send it to a model and get back a place to click. Then they do it all again. Every step costs image tokens and every click is a guess. It breaks whenever the layout changes. But the page already knows what it can do. So OpenAX flips it. Our ax-kit implementation follows the WebMCP draft from the W3C Web Machine Learning Community Group. Pages list the tools they offer and the browser runs the calls in your own tab. Nothing to read from pixels. Nothing to guess. Being usable by agents stops being extra work and starts being part of the page.