I am wondering for all those of you who have built chatbots in n8n or those that at storing vectorized data into a vector db like Qdrant, Pinecode, PGVector etc…
Have you found a way to update your data in the Vector store without knowing the actually vectors or ID’s of the chunks. In Pinecone, I am using a custom namespace and in Qdrant I am using collections to organize my documents. Only way that I have found to keep things updated is to delete a file before uploading the updated version (which seems stupid and cumbersome).
I think its great that we can upload documents into the stores, but the upsert or update functionality seems to elude me.
My goal is to keep the db’s clean with accurate data and I cane seem to figure this one out.
as far as I know this is the common way with upserting vector stores. You basically want to use metadata as much as possible to granulate the chunks, in order to minimize what is being deleted and re-inserted when updating.
Our community member @octionic made a pretty neat Youtube video about that if you want to check it out (he uses Supabase though but the concept is the same)
