A beginner question😅. When we use vector DB for RAG, if we update the content in the vector DB (e.g., adding new text files, replacing an existing file with updated content), do we need to re-create the vector embeddings from scratch?

You can just delete the corresponding records and reindex them.

It's an important question! You essentially are dealing with versioning and there isn't one single answer. You might want to add the

created_at

timestamp in the metadata, and do custom filtering of that at retrieval time, or you might only care about the latest in which case you should write the new record then delete the old one. If you aren't careful, you'll end up with dupes or close-dupes at retrieval time, so you're right to think about this!