Pinecone: Managed Vector Database for AI Applications

Getting Started with Pinecone

Install the Python SDK and create your first serverless index in under five minutes.

# Install the Pinecone Python SDK
pip install pinecone

# Initialize the client
from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key="YOUR_API_KEY")

# Create a serverless index (dimension must match your embedding model)
pc.create_index(
    name="my-rag-index",
    dimension=1536,           # OpenAI text-embedding-3-small default
    metric="cosine",          # cosine | dotproduct | euclidean
    spec=ServerlessSpec(
        cloud="aws",
        region="us-east-1"
    )
)

index = pc.Index("my-rag-index")
print(index.describe_index_stats())

Upsert, Query, and Fetch Operations

The three core operations: upsert vectors with metadata, query by similarity, and fetch by ID.

import openai

client = openai.OpenAI()

# Generate embeddings
def embed(texts):
    resp = client.embeddings.create(
        model="text-embedding-3-small",
        input=texts
    )
    return [r.embedding for r in resp.data]

# Upsert vectors with metadata
docs = [
    {"id": "doc-1", "text": "RAG systems retrieve context before generation", "category": "ai"},
    {"id": "doc-2", "text": "Pinecone supports real-time upserts", "category": "database"},
]
embeddings = embed([d["text"] for d in docs])

vectors = [
    {"id": d["id"], "values": emb, "metadata": {"text": d["text"], "category": d["category"]}}
    for d, emb in zip(docs, embeddings)
]
index.upsert(vectors=vectors, namespace="production")

# Query: find top-3 similar vectors with metadata filter
query_emb = embed(["how does retrieval augmented generation work?"])[0]
results = index.query(
    vector=query_emb,
    top_k=3,
    namespace="production",
    filter={"category": {"$eq": "ai"}},   # metadata filter
    include_metadata=True
)

for match in results.matches:
    print(f"{match.score:.4f} | {match.metadata['text']}")

# Fetch by ID
fetched = index.fetch(ids=["doc-1"], namespace="production")
print(fetched.vectors["doc-1"].metadata)

Namespace-Based Multi-Tenancy

Use namespaces to isolate data between tenants or environments within a single index.

# Upsert into separate namespaces per tenant
index.upsert(vectors=tenant_a_vectors, namespace="tenant-a")
index.upsert(vectors=tenant_b_vectors, namespace="tenant-b")

# Query is scoped to a single namespace — no cross-tenant leakage
results_a = index.query(vector=query_vec, top_k=5, namespace="tenant-a")
results_b = index.query(vector=query_vec, top_k=5, namespace="tenant-b")

# Delete all vectors for a tenant (data isolation for GDPR etc.)
index.delete(delete_all=True, namespace="tenant-a")

# Stats per namespace
stats = index.describe_index_stats()
for ns, ns_stats in stats.namespaces.items():
    print(f"{ns}: {ns_stats.vector_count} vectors")

Pinecone vs Weaviate vs Qdrant vs pgvector