Avoiding wrong results in conversationl-ReACT-description bot with zep long term memory

[axen2u]

I'm creating a langchain chatbot (Conversationl-ReACT-description) with zep long term memory and that bot has access to real-time data. How do I prevent use the memory when same query asked again few minutes later? The problem is when a tool produced an error or irrelevant result memory will give it agian and agian. Is there any workaround for that?

[danielchalef]

You'll want to think through how to separate your agent's task planning and execution from output. i.e. only store output and interactions with the user in Zep.

I suggest asking the LangChain Discord community for assistance if you get stuck.