Vector Databases Explained: Pinecone, Chroma & pgvector

Why Regular Databases Can't Do This

Traditional databases are great at exact matches: find all users where age = 30, or full-text search for documents containing "machine learning". But they can't answer questions like "find me documents that are semantically similar to this query" — because they don't understand meaning.

Vector databases solve this by storing data as high-dimensional vectors (embeddings) and enabling fast approximate nearest neighbor (ANN) search — finding the most similar vectors to a query vector in milliseconds, even across millions of records.

How Embeddings Capture Meaning

An embedding model converts text (or images, audio, etc.) into a fixed-size vector of numbers. Similar content produces similar vectors — measured by cosine similarity or dot product.

ANN Algorithms: How Fast Search Works

Brute-force search (compare query to every vector) is too slow at scale. Vector databases use ANN algorithms:

Pinecone: Managed Vector Database

import { Pinecone } from '@pinecone-database/pinecone';

const pc = new Pinecone({ apiKey: process.env.PINECONE_API_KEY });
const index = pc.index('my-index');

// Upsert vectors
await index.upsert([
  { id: 'doc1', values: [0.1, 0.2, ...], metadata: { text: 'Hello world', source: 'docs' } },
  { id: 'doc2', values: [0.3, 0.4, ...], metadata: { text: 'AI is amazing', source: 'blog' } },
]);

// Query for similar vectors
const results = await index.query({
  vector: queryEmbedding,  // Your query converted to embedding
  topK: 5,
  includeMetadata: true,
  filter: { source: { $eq: 'docs' } },  // Optional metadata filter
});

results.matches.forEach(match => {
  console.log(match.score, match.metadata.text);
});

Chroma: Local Development

import { ChromaClient, OpenAIEmbeddingFunction } from 'chromadb';

const client = new ChromaClient(); // Runs locally by default

const embedder = new OpenAIEmbeddingFunction({
  openai_api_key: process.env.OPENAI_API_KEY,
  openai_model: 'text-embedding-3-small',
});

const collection = await client.getOrCreateCollection({
  name: 'my-docs',
  embeddingFunction: embedder,
});

// Add documents (Chroma handles embedding automatically)
await collection.add({
  ids: ['doc1', 'doc2'],
  documents: ['Hello world', 'AI is amazing'],
  metadatas: [{ source: 'docs' }, { source: 'blog' }],
});

// Query by text (Chroma embeds the query automatically)
const results = await collection.query({
  queryTexts: ['What is artificial intelligence?'],
  nResults: 3,
});

pgvector: If You Already Use PostgreSQL

-- Enable the extension
CREATE EXTENSION IF NOT EXISTS vector;

-- Create table with vector column
CREATE TABLE documents (
  id SERIAL PRIMARY KEY,
  content TEXT,
  embedding vector(1536)  -- OpenAI text-embedding-3-small dimension
);

-- Insert with embedding
INSERT INTO documents (content, embedding)
VALUES ('Hello world', '[0.1, 0.2, ...]'::vector);

-- Find 5 most similar documents (cosine similarity)
SELECT content, 1 - (embedding <=> '[0.15, 0.25, ...]'::vector) AS similarity
FROM documents
ORDER BY embedding <=> '[0.15, 0.25, ...]'::vector
LIMIT 5;

-- Create HNSW index for fast search
CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops);

Choosing the Right Vector Database