Skip to main content

Pinecone

Pinecone is a managed, hosted vector database — no infrastructure to run, but the index itself is a billed cloud resource rather than a free local file.

from langchain_pinecone import PineconeVectorStore
from pinecone import Pinecone

pc = Pinecone(api_key="...")
index = pc.Index("example-index")

vector_store = PineconeVectorStore(embedding=embeddings, index=index)
vector_store.add_documents(documents, namespace="tenant-a")

Index creation and namespaces​

An index is created once (via the Pinecone console or API) with a fixed dimension and similarity metric — both must match the embedding model you plan to use, since a Pinecone index isn't automatically compatible with an arbitrary model. Namespaces partition a single index into isolated sub-collections, which is the usual pattern for multi-tenant applications instead of creating one index per tenant.

danger

The index dimension is set at creation and cannot be changed. If it doesn't match the embedding model's output size, every write or query fails outright — confirm the dimension before creating the index, not after.

Cost model​

Pinecone bills per index-hour (roughly, capacity reserved), not per query — an idle index with low traffic still costs the same as a busy one at the same pod size. That's the opposite of a pay-per-query API and changes the calculus for spinning up throwaway indexes during development.

See also​

  • Overview — the shared VectorStore interface.
  • Comparison — Pinecone against the alternatives.