pgvector vs Pinecone
· Source: original
Postgres with pgvector vs Pinecone: 200–300 thousand vectors don't require a separate vector DB
🧨
A startup was building a RAG system for legal documents and decided to move off Postgres to Pinecone. The volume of vectors was 200–300 thousand. The migration was approved based on a "general feeling" that Postgres isn't suitable for vectors. Without tests. They never tried pgvector at all.
Two years ago, the claim that "Postgres can't handle vectors" was true. Now pgvector covers vector search at such volumes — a separate vector DB isn't needed.
After a reference to the pgvector documentation, the solution was working within a week. According to the tech lead, it worked "even faster than they thought." A month later, the startup launched production on a single Postgres — without Pinecone and without a separate vector database.
They had already allocated 3 months of infrastructure work for the migration. Instead of three months — a week to a working solution and a month to production. The author of the analysis on Habr has observed this scenario at least 5 times over the past couple of years in various forms. The call with the tech lead of that very startup took place a couple of months ago, and the article was published on September 12, 2026