I've found that it's possible to quite easily drop a document with 200 000 tokens into a chat, which may incur rapidly escalating API costs if a user has a long thread in such a chat.
It might be worthwhile to pre-process and detect/estimate the number of tokens and warn users if it exceeds a certain threshold, e.g. "Your input contains more than 50 000 tokens, do you wish to proceed?". Perhaps with some disclaiming or notifiying about fair use limits. (And an option to disable warnings in Settings)
I'm imagining some less savvy users might unknowingly drop some documents the size of small books into chats and quickly end up costing Kagi a ton in API costs. E.g. someone thinks its a good idea to do some book reports with Assistant and spends 30 million tokens doing reports for 20 books.
If a certain token threshold is met, upon pressing enter or clicking send, a warning appears asking the user if they're sure they want to proceed, along with some information regarding fair use policy.
Could have a separate warning for a single message, and maybe a different warning if a thread is approaching a certain threshold.
