Feature request: load the complete conversation when a chat is opened, or prefetch the remaining history
ChatGPT on the web currently loads a conversation’s history in small instalments, and only when I scroll to the top of what is already loaded.
What I measured for a conversation of about 74 turns (a page about 281,000 px tall):
- On opening, one request fetches only the latest 10 items (GET /backend-api/conversations/?num_turns=10, 19 KB, 1.2 s).
- The rest arrives through 14 further requests of the form /backend-api/conversations//messages?before=&num_turns=10. Each is 3-27 KB (median 12 KB) and takes 0.6-1.4 s (about 13 s of request time in total). Each is triggered when I reach the top of the loaded history, so reading back through a long chat means repeated, visible loading pauses.
- The entire history is about 196 KB transferred. For comparison, the page downloads /backend-api/conversations/batch (236 KB) twice on every page load.
- Once loaded, only about 500 text blocks are kept mounted in that page. That windowed rendering is fine; this request is about when the data is fetched, not how it is drawn.
Please consider one of:
- loading the complete conversation when a chat is opened;
- prefetching the remaining history straight after the first view has rendered; or
- a preference such as “Preload complete conversations”.
The aim is not a larger model context or Memory. It is only to make the complete stored conversation immediately and continuously scrollable in the web interface.