something important i just figured out - if you have file names, don’t bother including the extensions when feeding through a prompt layer - it adds a significant amount of tokens as it treats each character or extension as a seperate word (logically should, but doesnt actually need to since it can contextualize that in single phrases like nextjs or typescript etc).
this should make your outputs a lot more reasonable token-wize if you’re working with something large. and then you can reappend the filenames later to the output it generates from the api call.
