Qdrant is also a good default choice, since it can work in-memory for development, with a hard drive for small deployments and also for "web scale" workloads.

As a principal eng, side-stepping a migration and having a good local dev experience is too good of a deal to pass up.

That being said, turbopuffer looks interesting. I will check it out. Hopefully their local dev experience is good

seems like a good rule of thumb to me! though i would perhaps lump "cost" into the "until it breaks" equation. even with decent perf, pg_vector's economics can be much worse, especially in multi-tenant scenarios where you need many small indexes (this is true of any vector db that builds indexes primarily on RAM/SSD)

I'd love to know how they compare versus MixedBread, what relative strengths each has. https://www.mixedbread.com/

I really really enjoy & learn a lot from the mixedbread blog. And they find good stuff to open source (although the product itself is closed). https://www.mixedbread.com/blog

I feel like there's a lot of overlap but also probably a lot of distinction too. Pretty new to this space of products though.

Often not a dealbreaker, actually! We can spin up new tpuf regions and procure dedicated interconnects to minimize latency to the on-prem network on request (and we have done this).

When you're operating at the 100B scale, you're pushing beyond the capacity that most on-prem setups can handle. Most orgs have no choice but to put a 100B workload into the nearest public cloud. (For smaller workloads, considerations are different, for sure.)

I don't know at these scales, but at the 1M-100M, we found switching from out-of-box embeddings to fine-tuning our embeddings gave less of a sting in the compression/recall trade-off . We had a 10-100X win here wrt comparable recall with better compression.

I'm not sure how that'd work with the binary quantization phase though. For example, we use Matroyska, and some of the bits matter way more than others, so that might be super painful.

So many missing details...

Different vector indexes have very different recall and even different parameters for each dramatically impact this.

HNSW can have very good recall even at high vector counts.

There's also the embedding model, whether you're quantizing, if it's pure rag vs hybrid bm25 / static word embeddings vs graph connections, whether you're reranking etc etc

the solution described in the blog post is currently in production at 100B vectors

(author here) The 92% mentioned in this post is showing recall@10 across all 100B vectors, calculated by comparing to the global top_k.

turbopuffer will also continuously monitor production recall at the per-shard level (or on-demand with https://turbopuffer.com/docs/recall). Perhaps counterintuitively, the global recall will actually be better than the per-shard recall if each shard is asked for its own, local top_k!