Thanks as always @pawlean
@heystu I fork the chat to clear context window. After like 10-15 messages the Mini Model goes squirrely and cost increases quite a bit. It’s just context management and extremely heavy babysitting of the model.
While I could use Claude or Max models and have bigger context, costs just go way too high, and they compound with added context of longer chats. Mini does most of what I need so I just mini-batch the dev and progress function-by-function after the main app layout is established. I use new chats for each function (ish), or when I hit 10-15 messages and the model goes wonky. Sometimes I have multiple chats per function if it struggles.
On a cost basis I’m probably saving 98% vs Claude or Max in a 40-60 message context window via this method with Mini. I’m sure Vercel doesn’t like that, but hey, thems the breaks. 
A full scale app could be 1,000 to 1,500 messages, or more. Depends. If we could clear context another way, that would be another method to reduce the number of chats, but ya. I don’t use GitHub, all local on Vercel for now. Thanks again man!