Thanks, but the index being in-memory makes it unsuitable for large data sets :/

There is a way of running disk-backed FAISS indexed that don't all fit in memory but I've not quite figured out how to do that yet: https://github.com/facebookresearch/faiss/issues/2675

OpenSearch K-NN plugin supports FAISS and it's disk based:

https://opensearch.org/docs/latest/search-plugins/knn/index/

OpenSearch looks like the best so far, all my requirements combined!

Can you say more? Usually projects that gravitate to SQLlite are not those that require massive scale and a FAISS index of a few GB covers a lot of documents.

My dataset is going to be around 10M documents. With OpenAI embeddings, that will be around 62GB. AFAIK SQLite should be able to handle that size, but I haven't tried.

This is not going to be my primary DB. I would update this maybe once a day and the update doesn't have to be super fast.

you might check out some vector databases:

https://milvus.io/

AND

pinecone.io

there are others too