Checking similarity

Yes! If you do something like index.query(..., response_mode="no_text") it will only fetch the source nodes and not call the LLM

You might also be interested in the similarity filtering option

index.query(..., similarity_cutoff=0.5)

But I'm not sure what the behavior is if all nodes get filtered out 🤔

I'll test if, an set it to 0.0 🤣

yes, I think cutoff might be what I need

maybe it should be in the demonstration or higher up in the docs?

seems like a pretty valid use case: Customers can ask about companies products, not about holidays in the Bahamas

I'll poke about in the code to see exactly what they both do

Sounds good! 👍

hmmmmm, I only ever get one item in response.source_nodes

is that expected behaviour for the default mode?

Yup! You can increase this though

index.query(..., similarity_top_k=4)

Does it will call LLM when I use index.query(..., similarity_cutoff=0.5) and not specify response_mode="no_text"?

It won't call the LLM, only the embeddings model (assuming you have a vector index)