Hello,
I’ve created a simple chatbot using an AI agent.
The bot accesses fixed values from a vector store and provides responses based on them.
However, I’ve run into a few problems that I haven’t been able to solve yet. Maybe someone here can help?
-
Topic memory issue:
At the beginning, the chat correctly asks which topic the user wants to discuss.
However, after a while, it asks for the topic again — which it shouldn’t.
It should stay within the selected topic.
The simple memory is set to a context window length of 45, which should be sufficient. -
Decreasing accuracy over time:
The more questions I ask, the less accurate the answers become.
It seems like the “memory” gets cluttered with previous questions and answers.
Is there a way to make it forget earlier information or clear the context periodically? -
Topic-routing AI agents:
As mentioned before, the chat starts by asking for the topic.
I’d like to build a system where one AI agent handles only that part — identifying the topic —
and then routes the conversation to another specialized AI agent.
For example, Agent 1 asks for the topic, finds the matching one, and then forwards the user to Agent “Sales”.
Is that possible?