interception

"The architecture of Fauxmium is relatively straightforward (although there is more complexity in my repository as I try to stream responses). You launch a browser via Puppeteer and set it up to intercept all requests. When a request is made, you send the URL to an LLM (along with a prompt to help it generate content), and it generates HTML or images, which are then returned to the browser."

"So I built a proof of concept off the back of fauxmium called interceptium [ code]. You launch Chrome for Testing via Puppeteer and set it up to intercept every request. Then, when a request is made, you decide if you want to handle the request or let it go to the network. If you want to handle it, you have the chance to change the request (you might want to automatically generate post-data, for example)."

An approach uses Puppeteer to launch a browser and intercept all network requests so an LLM can participate in generating or altering content. When a request occurs, the system can either forward it to the network or handle it locally after modifying request data, including automatically generating POST bodies. Responses from the network can be passed into an LLM, which synthesizes HTML or images to return to the browser. A proof-of-concept implements per-request decision logic and end-to-end LLM involvement at multiple points of the request lifecycle, enabling simulated or dynamically generated web content delivered directly to the browser.

Read at AI Focus