LLM is flagging "Yes/Sure" follow-ups as spam. How to bridge memory?

I'm building a booking bot. I have a Basic LLM Chain acting as a "Bodyguard" to filter intents. If it's a booking, it goes to an AI Agent (with Google Calendar tools); otherwise, it goes to a Spam branch.

The problem is when the AI Agent asks "Would you like to book 3 PM?", the user replies "Yes." Because the Bodyguard node is stateless/has no memory, it sees "Yes" in isolation, decides it's not a booking request, and routes it to the Spam branch. I need the Bodyguard to recognize that "Yes" is a continuation of the previous booking conversation. Anyone know of a solution?

3 comments