Each retrieval step requires it own API calls: – For SQL, you need to call an LLM to generate a query
– For vectorstore, you need to create an embedding And this is just for two retrieval sources – but it's easy to imagine extending this to more.

→ View original post on X — @hwchase17,