Pinecone — Managed Vector Database for Production AI

What it is

Pinecone is a fully managed vector database built for production AI applications. It stores vector embeddings and provides fast similarity search for use cases like semantic search, recommendation systems, and retrieval-augmented generation (RAG). Pinecone handles scaling, indexing, and infrastructure so you focus on your application logic.

Pinecone is designed for AI engineers and product teams building search, recommendation, or RAG features who need a production-ready vector store without managing infrastructure.

How it saves time or tokens

Self-hosting a vector database (Milvus, Weaviate, Qdrant) requires provisioning servers, managing indexes, tuning performance, and handling scaling. Pinecone eliminates all operational overhead. You create an index, upsert vectors, and query, all through a simple SDK. The serverless architecture scales automatically based on usage, and you pay only for what you store and query. For RAG applications, Pinecone's low-latency retrieval means you can fetch relevant context quickly, reducing the need for large context windows.

How to use

1. Install the Pinecone SDK:

pip install pinecone

1. Create a serverless index and upsert vectors:

from pinecone import Pinecone, ServerlessSpec

pc = Pinecone(api_key='your-api-key')

pc.create_index(
    name='docs',
    dimension=1536,
    metric='cosine',
    spec=ServerlessSpec(cloud='aws', region='us-east-1'),
)

index = pc.Index('docs')
index.upsert(vectors=[
    ('doc-1', [0.1, 0.2, ...], {'source': 'readme', 'topic': 'setup'}),
    ('doc-2', [0.3, 0.4, ...], {'source': 'api-docs', 'topic': 'auth'}),
])

1. Query with metadata filtering:

results = index.query(
    vector=[0.1, 0.2, ...],
    top_k=5,
    filter={'topic': {'$eq': 'auth'}},
    include_metadata=True,
)

Example

A RAG pipeline using Pinecone for context retrieval:

from openai import OpenAI
from pinecone import Pinecone

openai = OpenAI()
pc = Pinecone(api_key='...')
index = pc.Index('knowledge-base')

def ask(question: str) -> str:
    # Embed the question
    embedding = openai.embeddings.create(
        input=question, model='text-embedding-3-small'
    ).data[0].embedding

    # Retrieve relevant context
    results = index.query(vector=embedding, top_k=3, include_metadata=True)
    context = '\n'.join([m['metadata']['text'] for m in results['matches']])

    # Generate answer with context
    response = openai.chat.completions.create(
        model='gpt-4',
        messages=[
            {'role': 'system', 'content': f'Answer using this context:\n{context}'},
            {'role': 'user', 'content': question},
        ],
    )
    return response.choices[0].message.content

Related on TokRepo

• RAG tools — Browse retrieval-augmented generation tools
• Database tools — Explore database solutions for AI

Common pitfalls

• Using the wrong embedding dimension. Your index dimension must match the output dimension of your embedding model. OpenAI text-embedding-3-small produces 1536 dimensions; other models differ.
• Not using metadata filtering for hybrid search. Pinecone supports filtering by metadata fields alongside vector similarity. Without filters, you get pure similarity results which may include irrelevant matches.
• Creating too many indexes instead of using namespaces. Pinecone namespaces let you partition data within a single index, which is more cost-effective than creating separate indexes for each data source.