AI Tools · Const

pgvector

The RAG workflow uses pgvector with Postgres to store embeddings alongside chunked source content. Vector distance then supports similarity-based retrieval for AI applications.

How it works

The documented RAG workflow chunks source material, converts each chunk into an embedding, and stores the embedding with its text in PostgreSQL using the pgvector extension. Later, vector similarity can help retrieve relevant chunks for an AI application.

Syntax

pnpm db:migrate

Key points

Embeddings transform content into vectors that can be compared mathematically.
Vector proximity is used as a signal that two pieces of content have similar meaning.
Distance between vectors provides the similarity measure used for retrieval.
The workflow embeds each prepared chunk rather than processing unstructured source material as one unit.
The documented implementation combines PostgreSQL with pgvector for vector-backed storage.

Common mistakes

The database must support vector data; pgvector is used here because it adds vector support to Postgres.
Run the migration before relying on vector storage, because the extension is enabled during migration.

Example: Generate an embedding for vector storage

import { embed } from 'ai';
import { openai } from '@ai-sdk/openai';

const { embedding } = await embed({
  model: openai.embedding('text-embedding-3-small'),
  value: 'A guide to vector search',
});

console.log(embedding.length);

Creates an embedding that can be stored alongside source content for later vector similarity search.

When to use it

Retrieval-augmented generation

A documentation assistant needs to retrieve relevant passages before generating an answer.

Semantic search

An application needs semantic lookup over chunked text rather than exact keyword matching.

Embedding storage

A pipeline needs persistent storage for embeddings and their source text.

Interview questions

QHow does the documented RAG workflow use pgvector?

The workflow first splits source material into chunks, then embeds each chunk and stores the embedding together with the chunk. PostgreSQL is used with the pgvector extension to provide vector storage.

Follow-up: Why chunk documents instead of embedding an entire document? Smaller chunks allow retrieval to select more focused passages for downstream generation.

QWhat is the purpose of an embedding in this workflow?

An embedding represents text as a vector in a high-dimensional space. Similarity can be estimated from the distance between vectors, so nearby vectors represent semantically similar content.

Follow-up: What must the storage layer provide? It needs support for vector values and operations that compare their similarity or distance.

QWhat does the database migration do?

The migration command enables the pgvector extension and creates the resources table used by the tutorial. That table stores identifiers, content, and timestamps alongside the vector-related data defined by the schema.

Follow-up: What should be verified if vector queries fail immediately after setup? Confirm that the migration completed and that the extension is enabled in the target database.

Embeddings
Postgres
RAG Agent
resources schema

Source: RAG Agent Guide. Checked 2026-10-15.