Pinecone Quickstart

Pinecone is a managed vector database. Choose a supported serverless setup and verify account, region and billing requirements before creating resources.

#Prepare access

Create an appropriate key in the console↗ and record the intended cloud/region. Keep the actual key outside code.

bashCopy
pip install pinecone

#Small vector example

pythonCopy
import os
from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key=os.environ["PINECONE_API_KEY"])
name = "tutorial-vectors"
if name not in pc.list_indexes().names():
    pc.create_index(
        name=name,
        dimension=3,
        metric="cosine",
        spec=ServerlessSpec(cloud="aws", region="us-east-1"),
    )
# Wait for the index's ready state before the following operations.
index = pc.Index(name)
index.upsert(vectors=[
    ("id1", [0.1, 0.2, 0.3], {"topic": "AI"}),
    ("id2", [0.3, 0.4, 0.1], {"topic": "Biology"}),
])
print(index.query(
    vector=[0.1, 0.2, 0.3], top_k=2, include_metadata=True
))

These three-dimensional vectors are synthetic teaching data, not text embeddings or proof of retrieval quality. The original source's abbreviated 1536-dimensional example used ellipses; a real embedding workflow must provide complete vectors matching its model and index.

#Alternative ingestion

Supported integrated-embedding indexes can accept source text through a configured embedding model. This differs from supplying your own vectors; check the relevant create/upsert APIs.

#Constraints

Index dimensions must match the vectors. Namespaces can separate query scope, but application authorization still controls who may access each scope. Verify successful writes and read-back before adding more workflow layers.

#Next step

• Index management