Hello, i implemented the Multi Document Agents (V1). I recognized that the approach is not able to answer follow-up questions (e.g. "Give me more details of the first example you mentioned."). This is because the follow-up question is directly used for retrieving the tools in the custom object retriever and not the agents reasoning.

My solution for this would be to implement a CondenseQuestion mechanism in the custom object retriever. Do you know any other approaches?
Thanks for help 🙂