Are there any examples that explain how Quillit confines replies to only the content in my study?

Yes. Think of your latest trip to a fast food drive-through that uses AI to take your order and process it. If you ask it a question unrelated to the menu, it can't process the reply with external content because it is trained on the menu only. As Quillit uses only your content, it won't be helpful to you to ask it unrelated questions.