I find the kapa-gpt-index is sooo great, and I try to reproduce this effect on our document, but it didn't work out very well. I would like to ask how kapa organizes knowledge, and is there any best practice for this kind of technical documentation?

when i use gpt_index with create_llama_chat_agent, i find when i ask more questions, the token_counter query will not log, but when i just use index.query, the query log every times when ask question. Is there any different?