AI NewsInfrastructureAnnouncement
pg_pinecone deploys on Railway in one click, 15x smaller than pgvector
A Railway template runs Postgres 18 with pg_pinecone already set up. Pinecone says its index was 15x smaller than a pgvector HNSW index on 1 million vectors.
Image: Pinecone
Why it mattersA team that wants vector search in Postgres can release it from one page. The smaller index means the same server can hold more vectors.
Pinecone released a Railway template for pg_pinecone on 18 November 2026. Gavin Johnson wrote the announcement on Pinecone's blog. The template runs Postgres 18 with the pg_pinecone extension already installed and turned on. A user clicks Deploy on Railway, picks a name, and gets a Postgres database with vector search in it. No extension install step, no custom image to build.
What pg_pinecone is
pg_pinecone is Pinecone's vector search extension for Postgres. It is open source under the Apache 2.0 license, Pinecone says. The default engine is called native. It runs inside Postgres and needs no Pinecone account. A team that signs up for Railway can run vector search for free on its own database.
The extension has a second mode. A user can point it at Pinecone's managed service with an API key. The same SQL statements then run on Pinecone's cloud. A team that outgrows one Postgres server can move the search load across without a code rewrite.
The numbers Pinecone reports
The post gives one benchmark. Pinecone says that on 1 million vectors, "the pg_pinecone index was about 15 times smaller than a pgvector HNSW index". Pinecone also says the extension "delivers faster results at equivalent recall levels". And that "a smaller index lets the same Postgres service hold many more vectors for the same cost". The post gives no query latency number next to the size claim.
Three limits on the number, which Pinecone does not state. The 15x ratio is Pinecone's own reading of its own extension against a rival. There is no second lab's run of the same test. The index size depends on how you tune pgvector HNSW too. A team that cares will run the test on its own data.
What Railway brings
Railway is a hosting service that runs any Docker image from a template. The template that Pinecone just added is the shortest path from zero to a Postgres instance with pg_pinecone. Pinecone's post says the template is "Postgres 18 with pg_pinecone already installed and enabled". A small team can skip the Docker build and the install of a Postgres extension. Those are the two steps where a first-time user drops off.
What to read after this
A smaller index that is "faster at equivalent recall" is a strong claim. Open the pg_pinecone repository and run the benchmark the extension comes with, on a sample of your own vectors. Then compare it to pgvector HNSW with the same recall target and the same machine. The right choice is still the one your own data picks, even when Pinecone's number looks good.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

