Vector Database

Vector databases store embedding vectors alongside metadata (like file paths, function names, or code snippets). When you search, your query is converted to an embedding vector, and the database uses approximate nearest neighbor (ANN) algorithms to quickly find the most similar stored vectors. These algorithms (like HNSW or IVF) trade a small amount of accuracy for massive speed gains, making it possible to search millions of vectors in milliseconds.

AI coding tools that index your codebase use vector databases under the hood. When Cursor or Cline indexes your project, it creates embedding vectors for code chunks and stores them in a local vector database. When you ask a question, the tool converts your query to an embedding, searches the vector database for relevant code, and passes those code chunks to the LLM as context. This is the "retrieval" step in RAG that makes AI tools aware of your specific codebase.

// Using a vector database for code search (Chroma example)
import { ChromaClient } from "chromadb";

const client = new ChromaClient();
const collection = await client.createCollection({ name: "codebase" });

// Index code chunks
await collection.add({
  ids: ["auth-1", "auth-2", "payment-1"],
  documents: [
    "function validateToken(jwt) { ... }",
    "function refreshSession(userId) { ... }",
    "function processPayment(amount, card) { ... }"
  ],
  metadatas: [
    { file: "src/auth.ts", line: 15 },
    { file: "src/auth.ts", line: 42 },
    { file: "src/payment.ts", line: 8 }
  ]
});

// Semantic search — finds auth-related code
const results = await collection.query({
  queryTexts: ["how does user authentication work?"],
  nResults: 5
});

If you are building AI-powered features on top of your codebase, start with Chroma for prototyping and pgvector if you already use PostgreSQL. Migrate to Pinecone or Qdrant when you need production-grade performance and scaling.