Second this. It seems bizarre to have to scroll all the way back up to copy it.
- I was using a custom LLM based on Opus for a writing task in assistant v2.
- I kept the chat saved, then hopped onto a different chat with Sonnet.
- When I went back to the Opus chat, it started using Sonnet. It also started using internet searches, even though they were turned off. I also noticed the token count going crazy, using the info button. After a lengthy chat with Claude, involving detailed text analysis, it was at around 100k. Then, after getting kicked over to Sonnet and asking a single question, the count was over 300K.
The LLM should stay on the specified model. The token count should accurately reflect the tokens used.
It would be nice to have a 'number of sub results' option in advanced settings. Or a 'see more / fewer sub results' option or something. Just my thoughts.
For some searches (noticed especially on mobile) Kagi can pull a lot of results from one site. This is one fairly extreme example of it - you can't see any other sites without scrolling down, which isn't a great experience (the search term I used here was 'zen and rebirth'). It would be nice to be able to tweak the number of pages pulled from sites in cases like this, or even turn that feature off so Kagi only shows the main site. I've toggled on/off grouping but that just shows all the links separately (unless I've missed something here?)
