Vector Databases: Pinecone & Weaviate

Introduction

A vector database is what makes retrieval-augmented generation possible at scale — it stores embeddings (numeric representations of meaning, covered fully in lesson 14) and can find the most similar ones to a query in milliseconds, even across millions of records. Pinecone and Weaviate are two of the most widely used managed options.

A Real-Life Analogy First

Imagine a giant library with no organization at all — every book dumped in one room, in no particular order. Finding a book about "cooking Italian food" means checking every single book by hand. Now imagine that same library organized by a librarian who has grouped every book by how similar its subject matter is to every other book's, so "Italian cooking," "pasta recipes," and "Mediterranean cuisine" all sit physically near each other on the shelf, even though they don't share the exact same words in their titles. A vector database is that second, meaning-organized library, but for pieces of text (or images) instead of books, and searchable in milliseconds instead of by walking the aisles.

What a Vector Database Actually Stores

Each record is a vector — a list of a few hundred to a few thousand numbers produced by an embedding model — plus optional metadata (like the original text or a source URL). A query is itself embedded into the same kind of vector, and the database returns the stored vectors whose numbers are mathematically closest to it, which corresponds to "most similar in meaning."

Pinecone: Managed Vector Search

Use case: Pinecone is a fully managed, serverless vector database — no infrastructure to run yourself, built specifically and only for vector search, with strong performance at large scale.

pip install pinecone

Pinecone in Action

from pinecone import Pinecone

pc = Pinecone()  # reads PINECONE_API_KEY from the environment
index = pc.Index("product-docs")

# Each vector: an id, its embedding, and optional metadata
index.upsert(vectors=[
    {"id": "doc-1", "values": [0.02, 0.91, -0.13, 0.44], "metadata": {"text": "Refunds are processed within 5-7 business days."}},
    {"id": "doc-2", "values": [0.88, -0.12, 0.30, 0.09], "metadata": {"text": "Shipping takes 3-5 business days domestically."}},
])

results = index.query(vector=[0.03, 0.89, -0.10, 0.41], top_k=1, include_metadata=True)
print(results.matches[0].metadata["text"])

Weaviate: Open-Source With Managed Hosting

Use case: Weaviate is an open-source vector database you can self-host or run through Weaviate's managed cloud — a good fit when you want the option to move off a fully proprietary managed service later, or need it to run inside your own infrastructure for compliance reasons.

pip install weaviate-client

import weaviate

client = weaviate.connect_to_weaviate_cloud(
    cluster_url="https://your-cluster.weaviate.network",
    auth_credentials=weaviate.auth.AuthApiKey("your-api-key"),
)

docs = client.collections.get("ProductDocs")
response = docs.query.near_text(query="how long do refunds take", limit=1)

print(response.objects[0].properties["text"])
client.close()

Pinecone vs Weaviate

Where You've Already Seen This Idea

Similarity search is not a brand-new concept invented for AI — you have already relied on it, often without realizing.

Common Mistakes

Best Practices

Frequently Asked Questions

Key Takeaways

Summary

Pinecone and Weaviate both solve the same core problem — fast similarity search over embeddings at scale — with different tradeoffs around hosting and openness. The next lesson covers lighter-weight, local-first alternatives.