How to Intermediate · 3 min read

How to use Pinecone with LlamaIndex

Q: How to use Pinecone with LlamaIndex

Use the pinecone Python SDK to create and manage a vector index, then connect it with LlamaIndex by initializing a PineconeVectorStore as the vector store backend. This enables efficient similarity search and retrieval within LlamaIndex using Pinecone's managed vector database.

Quick answer

Use the pinecone Python SDK to create and manage a vector index, then connect it with LlamaIndex by initializing a PineconeVectorStore as the vector store backend. This enables efficient similarity search and retrieval within LlamaIndex using Pinecone's managed vector database.

PREREQUISITES

Python 3.8+
Pinecone API key (free tier available)
pip install pinecone-client llama-index
Set environment variables PINECONE_API_KEY

Setup

Install the required Python packages and set your Pinecone API key environment variable.

Install packages: pip install pinecone-client llama-index
Export environment variable in your shell:
export PINECONE_API_KEY='your-pinecone-api-key'

bash

pip install pinecone-client llama-index

Step by step

This example demonstrates creating a Pinecone index, initializing LlamaIndex with Pinecone as the vector store, inserting documents, and querying for similar content.

python

import os
from pinecone import Pinecone
from llama_index import GPTVectorStoreIndex, SimpleDirectoryReader, PineconeVectorStore

# Initialize Pinecone client
pc = Pinecone(api_key=os.environ["PINECONE_API_KEY"])

# Create Pinecone index if it doesn't exist
index_name = "llamaindex-example"
if index_name not in pc.list_indexes():
    pc.create_index(index_name, dimension=1536, metric="cosine")

# Connect to Pinecone index
pinecone_index = pc.Index(index_name)

# Initialize PineconeVectorStore for LlamaIndex
vector_store = PineconeVectorStore(pinecone_index)

# Load documents (replace with your own data source)
docs = SimpleDirectoryReader("./data").load_data()

# Create LlamaIndex with Pinecone vector store
index = GPTVectorStoreIndex(docs, vector_store=vector_store)

# Query the index
query_str = "What is the capital of France?"
response = index.query(query_str)
print("Query response:", response.response)

output

Query response: Paris is the capital of France.

Common variations

Use different embedding models by passing a custom embed_model to GPTVectorStoreIndex.
Use async Pinecone client with pinecone.AsyncIndex for asynchronous operations.
Switch to other vector stores like FAISS or Chroma by changing the vector store initialization.

python

from llama_index import LangchainEmbedding
from langchain_openai import OpenAI

# Example: Using OpenAI embeddings with LlamaIndex and Pinecone
embed_model = LangchainEmbedding(OpenAI(model_name="gpt-4o"))
index = GPTVectorStoreIndex(docs, vector_store=vector_store, embed_model=embed_model)

Troubleshooting

If you see IndexNotFoundError, ensure the Pinecone index is created and the name matches.
For authentication errors, verify your PINECONE_API_KEY environment variable is set correctly.
If queries return empty results, check that documents were properly indexed and embeddings generated.

✅

Key Takeaways

Initialize Pinecone client with API key before use.
Use PineconeVectorStore to integrate Pinecone with LlamaIndex seamlessly.
Create the Pinecone index with the correct vector dimension matching your embeddings.
Load and index documents before querying to get meaningful results.
Verify environment variables and index existence to avoid common errors.

Verified 2026-04 · gpt-4o

Verify ↗