how can i extract a pydantic object(API parmas) from a user query?

Let's say I want to build muti-document agents from a vector database like Pinecone. Do I need to have a separate index for every document? I am aware that Pinecone allows the storage of vectors under different namespaces in a single index which can then be retrieved . However, I do not see that implementation in the Pinecone class in the llama-index. Or is there a different way to do it? Any suggestions?

I was wondering when we store some extracted metadata in the nodes, during query time how does the query engine look for that metadata and if it can be customized.
Ref- https://docs.llamaindex.ai/en/latest/examples/metadata_extraction/EntityExtractionClimate.html#entity-metadata-extraction