[fabceolin]

Clean context for each iteration will make the LLM give your better results. Using LLM loop you will full the context faster degrading the LLM responses.