How does RAG retrieval work across multiple queries in a single conversation?

Questions
1

I’m using the RAG / knowledge base feature, and I’m curious about how content is fetched from the knowledge base during a multi-turn conversation. Specifically, does Pickaxe only retrieve relevant chunks on the very first query, or does it run additional retrieval calls for subsequent follow-up queries in the same interaction?

I’m concerned about accuracy if the knowledge base is only queried once (as it feels like might be the case). How does the RAG pipeline handle continuity and context across multiple interactions?

Thanks in advance for any guidance or best practices you can share!

2 Likes
2

Pickaxe retrieves chunks on every query.

1 Like
3

Hi @transitionlevel you can check what chunks have been retrieved on every turn by clicking on the magnifier icon at the bottom of the response. You will notice that based on the question the chuncks vary.

image
image720×306 37.9 KB

Also the amount of chuncks and their relevance to the query can be modified in the knowledge settings.