Working with Pinecone (and other vectorstores): namespaces or seperate VS?

Hi guys

I have a general 'how to approach' things question re how best to structure vectorstores (Pinecone in this case):

I have a Pinecone Index (knowledge_base) with a namespace in it (called 'docs'). It's constituted of 20+ PDFs that were ingested via an n8n pipeline. All good on that regard.

The next step would be to query another handful of PDFs 'against' this index. I can run the same 'ingesting' flow for this query and now have 2 options:

Create a new vectorstore (query_1 etc) with these new info
Update the same vectorstore with namespace (query_1 etc)

I can see on n8n that I can dynamically set either the ID of the index or the namespace of the index so on both counts it should be arbitrary to setup a tool for the query itself; as the tool for the knowledge_base is static and doesn't change.

Which route would you suggest? Multiple indexes or one index with multiple namespaces?

1 comment