Hi.
Yes, you’re going to need extensive referencing in order for your cGPT to have ready access to the information you’re creating if it’s stored in PDFs.
I’ve found this starts by referencing a long-document in your Instructions. You can tell the model how to read that document and how it is organized, and it will get to the information you want it to faster.
However, there is a 20 document limit on the CustomGPT knowledge base, and longer documents become cumbersome quickly. So, you’ll need a well designed spreadsheet for relating information.
It is currently not easy to save game states, or even have enough of a context window for anything very meaningful.
With DungeonMasterGPT we’ve currently got it trained, and it finds information fairly quickly, but that document limit is causing problems. You can read more about how we’re dealing with it here.