Long chat performance

I find that after I chat for while, interactive performance can degrade. It would be really great if it scaled better so that it continued to respond to user input fluidly. The chat in question is 865 kB (that’s the size of the markdown export anyway) and many of the user messages contain an uploaded images (vision). When I started the chat, there was a short delay to do the inference of course, but otherwise it responded quickly. Now when I send a message the UI freezes and I can’t scroll the view, and I often see a “Page unresponsive” error.

Maybe this is because the UI is one big scrolling view? Would it help if it were broken into pages somehow? Also, if the software isn’t already doing so, would it help to summarize the earlier parts of the chat to reduce the amount of input data used for inference?

I’m running in Chrome on Windows 11, on an Intel Core i9-90900KF @ 3.70 GHz with 64 MB RAM.