VectorChord: Store 400k Vectors for $1 in PostgreSQL

This looks like an interesting project.<p>Though it&#x27;s worth noting that the license is AGPL. So if the idea is for this to take over for pgvecto.rs, it&#x27;s an important data point for those building SaaS products.<p>It will make pgvector the only permissively licensed option, given it has the same license as Postgres.

Could you talk about how updates are handled? My understanding is that IVF can struggle if you&#x27;re doing a lot of inserts&#x2F;updates after index creation, as the data needs to be incrementally re-clustered (or the entire index needs to be rebuilt) in order to ensure the clusters continue to reflect the shape of your data?

Awesome work! But aren&#x27;t the comparisons missing ScaNN [1, 2]? I think it&#x27;s the overall SOTA [3] at the moment regarding vector indexing.<p>[1] <a href="https:&#x2F;&#x2F;github.com&#x2F;google-research&#x2F;google-research&#x2F;tree&#x2F;master&#x2F;scann">https:&#x2F;&#x2F;github.com&#x2F;google-research&#x2F;google-research&#x2F;tree&#x2F;mast...</a><p>[2] Also available on something like AlloyDB on GCP: <a href="https:&#x2F;&#x2F;cloud.google.com&#x2F;alloydb&#x2F;docs&#x2F;ai&#x2F;store-index-query-vectors?resource=scann" rel="nofollow">https:&#x2F;&#x2F;cloud.google.com&#x2F;alloydb&#x2F;docs&#x2F;ai&#x2F;store-index-query-v...</a><p>[3] <a href="https:&#x2F;&#x2F;ann-benchmarks.com&#x2F;glove-100-angular_10_angular.html" rel="nofollow">https:&#x2F;&#x2F;ann-benchmarks.com&#x2F;glove-100-angular_10_angular.html</a><p>Disclaimer: Working for Google, but nowhere close to Databases.

The “external index build” idea seems pretty interesting. How does it work with updates to the underlying data (e.g., new embeddings being added)? For that matter, I guess, how do incremental updates to pgvector’s HNSW indexes work?

It would have been nice a comparison with pgvectorscale, which uses binary quantization and StreamingDiskANN.

The cost to store a static set of 400k 768-dimension vectors is also $1 a month on Datastax&#x27;s AstraDB. However, for that $1, AstraDB replicates the data 3x instead of storing it on a single machine.<p>Here is a link to the cost calculator. Note that the calculator includes cost of ingestion, but the article only mentions storage costs, not ingestion costs: <a href="https:&#x2F;&#x2F;www.datastax.com&#x2F;pricing&#x2F;vector-search?cloudProvider=aws&amp;cloudRegion=us-east-2&amp;nonVectorReadSizeBytes=1024&amp;nonVectorWriteSizeBytes=4096&amp;vectorDimensions=768&amp;vectorReadsPerMonth=0&amp;vectorWritesPerMonth=400000" rel="nofollow">https:&#x2F;&#x2F;www.datastax.com&#x2F;pricing&#x2F;vector-search?cloudProvider...</a><p>Disclaimer: I work on vectorsearch&#x2F;AstraDB at DataStax.

I am still waiting for a good pattern for using multivector embeddings like ColBert and ColPali in postgres. I get that its fun to optimize single vector stuff, but multivector is that happy middleground between single vector and reranker that seems to be only validated in specialized exotic search dbs like Vespa

Hey everyone! We’ve developed a new PostgreSQL extension that supports 400k vectors for just $1. Check it out!

Would you be willing to speculate on how VectorChord&#x27;s ingestion and query performance might compare to Elasticsearch&#x2F;OpenSearch for dense vector and sparse vector search use cases, particularly when dealing with larger full text data sets (&gt;5M records)?

Does this mean you won&#x27;t support pgvecto.rs anymore?

What dimension vectors are we talking here

In five pages of text, we never get to learn what a Vector is (in this context), why we’d want to store one in pgsql, or why it costs so much to store them compared to anything else you’d store there.<p>For an example of how you can communicate with domain experts, while still giving everyone else some form of clue as to what this hell you’re talking about, check out the link to the product that this thing claims to be a successor to:<p><a href="https:&#x2F;&#x2F;pgvecto.rs&#x2F;" rel="nofollow">https:&#x2F;&#x2F;pgvecto.rs&#x2F;</a><p>That starts off by telling us what it is and what it does.