Introducing pg_pinecone: Open-Source Vector Search Inside Postgres

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Introducing pg_pinecone: Open-Source Vector Search Inside Postgres · Channel: Pinecone · 01:38 · Watch on YouTube

Topics: Data, databases and embeddings, Open source and software ecosystems, RAG, search and information retrieval

Summary

PG Pinecone is an open-source Postgres extension based on the search engine Pinecone runs at scale. It offers a small index, fast builds, and accurate filtered or unfiltered vector search inside Postgres, using the same SQL interface without new infrastructure. The video demonstrates installing the extension, creating an index, and querying by tenant ID and embedding. It also notes that users can move search to Pinecone and add embeddings, keyword search, and re-ranking with an API key.

Key points

PG Pinecone is an open-source Postgres extension based on Pinecone’s search engine.
It is presented as an alternative to workarounds such as per-tenant indexes, limiting indexed data, overnight rebuilds, or larger instances.
It provides a small index, fast builds, and accurate filtered or unfiltered search.
Search runs inside the Postgres database and retains the same SQL interface without requiring new infrastructure.
Users can install it from a pre-built GitHub release or a Docker Hub image.
To get started, create the extension, use an existing vector column or add a new one, and create an index.
The example uses a tenant ID column to filter queries, then orders results by embedding.
The video says a command can move search to Pinecone while keeping the same queries and Postgres as the source of truth.
With an API key, users can also add embeddings, keyword search, and re-ranking.

Chapters

Start	Chapter	Summary
00:03	Introducing PG Pinecone	The video describes common Postgres vector-search workarounds and introduces PG Pinecone as an open-source extension for fast search inside the database.
00:34	Install and create an index	The presenter outlines installation, creating the extension, selecting a vector column, and indexing it for tenant-based filtering.
01:06	Query and get started	The demo runs a filtered SQL query, describes moving search to Pinecone and optional features, and invites viewers to try the extension.

Quotes

But you picked Postgres to keep things simple. — 00:03

So today we’re introducing PG Pinecone, an open-source Postgres extension based on the search engine that Pinecone runs at scale. — 00:03

You get a small index, fast builds, accurate filtered or unfiltered search all right inside your database. — 00:34

The same SQL interface right inside your Postgres database. — 01:06

Create the extension, add the index, and run your first filtered query. — 01:38

Concepts and entities

PG Pinecone, PostgreSQL, Pinecone, GitHub, Docker Hub

Full transcript (original language)

00:03 How many workarounds are you running to keep vector search fast in your Postgres database? Maybe you’ve got an index per tenant, or you’re capping how much data you index. Or you’ve tried rebuilding overnight and it doesn’t finish in time. Maybe you have a bigger instance just so your index fits in memory. But you picked Postgres to keep things simple. So today we’re introducing PG Pinecone, an open-source Postgres extension based on the search engine that Pinecone runs at scale. You get a small index, fast builds,

00:34 accurate filtered or unfiltered search all right inside your database. And you keep the same SQL interface and no new infrastructure. Getting started takes just a few minutes. Grab the pre-built release from GitHub or pull the image from Docker Hub. Then [music] in Postgres, create the extension. You can either add a new vector column or use one that you already have. >> [music] >> Then create an index on it. Here, I’m using a tenant ID column for filtering

01:06 by tenant. [music] And then query the way you always do. Filter by tenant ID and then order by embedding. >> [music] >> That’s it. The same SQL interface right inside your Postgres database. >> [music] >> When you all crawl one server, one command moves your search to Pinecone. >> [music] >> You’ve got the same queries and Postgres stays your source of truth. With an API key, you can also add embeddings, keyword search, and re-ranking. >> [music] >> PG Pinecone is open-source and ready to try today. Run it on a table you already

01:38 have. Create the extension, add the index, and run your first filtered query. >> [music] >> Grab the links below to get started.