We have 19 Json files full of data, in debug mode, before publishing everything works fine.
If we hit save, it says “error saving” and if we go to GPT after publishing and ask a question, it won’t search the knowledge base and asks the user to tell her the exact file name to search!
In fact, the “error saving” appears if you have 10 files. It can’t search 10 files. Maybe even less.
We can’t merge all the files into 1, because that would be too many tokens. We’re already using the maximum number of tokens per file.
We have these instructions
When a user asks a question, search all the files in your knowledge base and retrieve an answer.