It’s not a model issue. It’s something to do with caching. Cline costs much less with same model.
Discussion
nostr:nprofile1qqsqgc0uhmxycvm5gwvn944c7yfxnnxm0nyh8tt62zhrvtd3xkj8fhgprdmhxue69uhkwmr9v9ek7mnpw3hhytnyv4mz7un9d3shjqg4waehxw309aex2mrp0yhxgctdw4eju6t09uq3vamnwvaz7tmjv4kxz7fwd4hhxarj9ec82c30hyvdwq added caching a while back. I would guess what you're describing is the large context window of using the #MKStack template? Are you using that with cline?
So make the initial prompt full context and subsequent as new task - like cline. Why does it need full app context to fix a small bug?
I don’t know how cline works exactly but it seems to search for relevant files first and uses that at the context for token cost - and this seems to really keep costs down. I can prompt for like an hour on $10 whereas in Shakespeare that’s basically 1 or 2 subsequent prompts and your money is gone 🤣
I mean it’s great for a starter template but if you want to fix bugs or enhance it much further: you have no choice but to export somewhere. I use vscode/cline which saves a lot on costs but it would be nice if I didn’t have to - I think all of your users would be happier too.
nostr:npub1qny3tkh0acurzla8x3zy4nhrjz5zd8l9sy9jys09umwng00manysew95gx made an excellent point in his last conversation with nostr:npub1q3sle0kvfsehgsuexttt3ugjd8xdklxfwwkh559wxckmzddywnws6cd26p that he could monetize hosting. But you’d have a much easier time doing it if people could continue developing in shakespear and not have to export with no way to import (lose out on hosting revenue)
some of these things are coming with ACTII.