Pinecone Basics

Introduction

Pinecone is the most popular managed vector database. It abstracts away index tuning, scaling, and replication — you just upsert vectors and query.

Beginner analogy: The Stripe of vector databases — managed, simple, fast.

Understanding the topic

Core concepts to understand:

• Sign up → create an index (specify dimension + metric).
• Upsert vectors with metadata.
• Query: send a vector, get top-k matches.
• Auto-scales; serverless tier available.
• REST + gRPC + native SDKs (Node, Python).

Syntax reference

Visual workflow / architecture:

pinecone.index("docs").upsert([
  { id: "1", values: [0.1, ...], metadata: {url: "..."} },
])

pinecone.index("docs").query({
  vector: [0.12, ...],
  topK: 5,
  includeMetadata: true,
})

Real-world use

Perplexity, Notion, Gong, Glean and many YC AI startups run on Pinecone.

Best practices

• Use namespaces to isolate tenants.
• Store source URLs in metadata for citations.
• Use serverless tier for variable workloads.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. What is Pinecone?
• Q2. How do you query for top-k similar vectors?
• Q3. Why use namespaces?