I dont know how to do solve?

@RichardCray re-reading happens because your memory node dumps the full chat history into context every turn. Swap to Window Buffer Memory with contextWindowLength: 5 so only the last 5 messages replay. Wire it into the AI Agent’s aimemory input alongside your OpenAI Chat Model node. Then add a systemMessage like “Respond only to the latest patient message, do not summarize prior turns” — that stops it re-analyzing previous turns.