This post is a follow-up to my previous post on how to setup a local MySQL instance in docker.

RAG (Retrieval Augmented Generation) is quickly becoming the "Hello World" of AI apps. If you are working or playing with Large Language Models, you will no doubt need to create a RAG pipeline at some point. An important component of RAG is a vector database, and a popular option is pgvector - an open-source vector similarity search for Postgres. Here's how to quickly setup a local instance in a Docker container.

Pull and run the image

Pull the latest image from the docker repository. Replace 17 with your Postgres server version of choice.

docker pull pgvector/pgvector:pg17

Run the image, set the root user password, and expose the default Postgres port.

docker run -d --name <container_name> -e POSTGRES_PASSWORD=postgres -p 5432:5432 pgvector/pgvector:pg17

Create a db inside the container

With the Postgres server running, create a database inside the container.

docker exec -it <container_name> createdb -U postgres <database_name>

Connect to the database

Now we can connect to the database from our application and initialize the pgvector extension. I'll be using JavaScript. Setting up the entire application is outside the scope of this post, but you will need to install a couple dependencies:

pnpm add pg pgvector

Set a DATABASE_URL in your environment. I use a .env file. It should follow this format:

DATABASE_URL=postgresql://<pg_user>:<pg_password>@localhost:5432/<database_name>

For local development use @localhost, but if you are using something like docker-compose.yml and have named the service, you should use the name of the service e.g. @db.

In your application code, create the connection:

const pool = new pg.Pool({
  connectionString: process.env.DATABASE_URL,
});

Then, initialize pgvector and create a new table:

async function createStore() {
  // Initialize pgvector extension and create table if not exists
  await pool.query('CREATE EXTENSION IF NOT EXISTS vector');

  return {
    vectorStore: await PGVectorStore.initialize(embeddings, {
      postgresConnectionOptions: {
        connectionString: process.env.DATABASE_URL,
      },
      tableName: 'documents', // Default table name
    }),
  };
}

With the vectorStore setup, you can add content to it using vectorStore.addDocuments and query for context using vectorStore.similaritySearch.

That's it for this post. Maybe next time I will explore more specific uses of pgvector, and/or using it with Drizzle ORM! 👋