Not storing prompts but they absolutely can store reasoning traces, summarized or sanitized reasoning traces and the whole LLM output as well.