Yeah, given a prompt for a website like "Tumblr Clone for Ascii Art", the LLM will pretend to be a HTTP server.

All client side requests are directly fed into the LLM and it will respond with raw HTTP responses that are then forwarded back to the client.

There is no persistence between requests. So if you want something dynamic it is best to describe specific routes.

An example prompt would be:

A tumbler clone but for ASCII art. Please make it have a fully functional infinite scroll. For the infinite scroll all requests to '/scroll' should be return a batched lot ascii art text to be embedded. The format for the /scroll requests will be a raw ASCII art followed by <br> for batching. Try to keep requests to a minimum for the infinite scroll.