so the main feature of claude project is its ability to have internal knowledge and a set of custom instructions (which mostly affects how it answers questions) Isn't this basically just uploading the files to a vector db and attaching it to the llm ? Plus a simple system instruction which asks it to follow the provided custom instructions when answering, and use the vector database as long term memory ? Is it really this simple or am I missing anything ? Like what's the hype really about ?