Do you still need a vector database for RAG?
Posted by Applied AI Commons team (community team) on
Long context windows, hybrid keyword search and database extensions have changed the default advice on retrieval. Some teams now skip a dedicated vector store entirely, while others say it is still essential at scale. The right choice seems to depend on corpus size, update frequency and how precise answers need to be.
Some angles worth exploring: - Your corpus size and how often it changes - Whether you use keyword search, embeddings, or both, and how you rank results - How you measure retrieval quality rather than just answer quality - What made you add, or remove, a vector database
For your retrieval setup today, do you rely on a dedicated vector database, and what would make you change that decision?