LlamaIndex Pinecone Vector Store Integration

0.8.0 · active · verified Thu Apr 16

This library provides the integration for using Pinecone as a vector store backend within LlamaIndex applications. It enables storing and retrieving document embeddings in a Pinecone index for efficient semantic search and Retrieval-Augmented Generation (RAG). As of version 0.8.0, it supports LlamaIndex's modular architecture, requiring separate installation from the core LlamaIndex library. It follows a frequent release cadence, often aligning with LlamaIndex core updates.

Common errors

```
AttributeError: 'PineconeVectorStore' object has no attribute 'service_context'
```
cause `ServiceContext` was deprecated in LlamaIndex v0.10.0 and `PineconeVectorStore` no longer relies on it directly.

fix Remove any explicit usage of `service_context` when initializing `PineconeVectorStore` or `VectorStoreIndex`. Configure LLM and embedding models directly using `Settings` or by passing them as arguments to `VectorStoreIndex.from_documents()`.
```
pinecone.exceptions.PineconeException: The dimension of the vectors to be upserted (X) does not match the dimension of the index (Y).
```
cause The vector dimension generated by your embedding model does not match the dimension specified when creating the Pinecone index.

fix Ensure the `dimension` parameter in `pc.create_index()` matches the output dimension of your embedding model. For example, if using OpenAI's `text-embedding-ada-002`, the dimension should be 1536.
```
Index 'your-index-name' is not ready. Please wait a few seconds and try again.
```
cause The Pinecone index creation can take a short amount of time to become active and ready for operations.

fix Implement a retry mechanism with a short delay (e.g., `time.sleep(5)`) or check `pc.describe_index(index_name).status` before proceeding with upserts or queries.

Warnings

breaking LlamaIndex v0.10.0 introduced a major packaging refactor. Core components moved to `llama-index-core`, and integrations like `pinecone-vector-store` are now separate PyPI packages.
Fix: Ensure you `pip install llama-index-core` and `pip install llama-index-vector-stores-pinecone`. Update imports from `from llama_index import ...` to `from llama_index.core import ...` for core components and `from llama_index.vector_stores.pinecone import ...` for this integration. The `ServiceContext` abstraction has also been deprecated; configure LLMs/embeddings directly or use global settings.
gotcha Pinecone index dimensions must match the embedding model's output dimension. Mismatched dimensions will lead to errors during upsert operations.
Fix: When creating a Pinecone index (e.g., `pc.create_index`), ensure the `dimension` parameter matches the output dimension of your chosen embedding model (e.g., 1536 for OpenAI's `text-embedding-ada-002`). Refer to your embedding model's documentation for the correct dimension.
gotcha Inconsistent or empty query results from Pinecone often stem from issues with API keys, index state, overly restrictive filters, or problems during document ingestion.
Fix: Verify `PINECONE_API_KEY` is correct and has access to the specified index. Check if documents were successfully added to Pinecone. Review any `MetadataFilters` applied during querying to ensure they are not inadvertently excluding relevant results. For persistent issues, inspect Pinecone's dashboard to confirm index content and health.
gotcha Compatibility issues can arise when `pinecone-client` is installed alongside other libraries that also depend on it (e.g., `langchain-pinecone`), leading to version downgrades or conflicts.
Fix: Use `pip install --upgrade` for specific packages to force the desired versions, or install `pinecone-client` directly with a version constraint that satisfies all dependencies (e.g., `pinecone-client>=4.0.0,<5.0.0`). Check the dependency requirements of all involved libraries.

Install

pip install llama-index-vector-stores-pinecone Install only Pinecone integration
pip install llama-index llama-index-vector-stores-pinecone pinecone-client openai Install with LlamaIndex core, Pinecone client, and OpenAI (common setup)

Imports

PineconeVectorStore

from llama_index.vector_stores.pinecone import PineconeVectorStore

VectorStoreIndex

from llama_index.core import VectorStoreIndex

SimpleDirectoryReader

from llama_index.core import SimpleDirectoryReader

StorageContext

from llama_index.core import StorageContext

Pinecone
wrong:
```
from llama_index.vector_stores.pinecone import Pinecone
```
correct:
```
from pinecone import Pinecone
```
The Pinecone client itself is imported directly from the `pinecone` package, not from the LlamaIndex integration.

Quickstart

This quickstart demonstrates how to set up a Pinecone index, initialize `PineconeVectorStore`, load documents using `SimpleDirectoryReader`, and build a `VectorStoreIndex` for querying. It assumes `PINECONE_API_KEY` and `OPENAI_API_KEY` are set as environment variables.

import os
from pinecone import Pinecone, ServerlessSpec
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, StorageContext
from llama_index.vector_stores.pinecone import PineconeVectorStore

# Set your API keys (replace with actual keys or use environment variables)
os.environ['PINECONE_API_KEY'] = os.environ.get('PINECONE_API_KEY', 'YOUR_PINECONE_API_KEY')
os.environ['OPENAI_API_KEY'] = os.environ.get('OPENAI_API_KEY', 'YOUR_OPENAI_API_KEY')

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

index_name = "quickstart-index"
if index_name not in pc.list_indexes().names():
    pc.create_index(
        name=index_name,
        dimension=1536, # Dimension for OpenAI's text-embedding-ada-002
        metric="cosine",
        spec=ServerlessSpec(cloud="aws", region="us-west-2")
    )

pinecone_index = pc.Index(index_name)

# Initialize PineconeVectorStore
vector_store = PineconeVectorStore(pinecone_index=pinecone_index)

# Load documents (create a 'data' directory with text files or adjust path)
try:
    documents = SimpleDirectoryReader(input_dir="./data").load_data()
except FileNotFoundError:
    print("Please create a 'data' directory and add some text files, or modify SimpleDirectoryReader path.")
    documents = []

if documents:
    # Set up StorageContext
    storage_context = StorageContext.from_defaults(vector_store=vector_store)

    # Create VectorStoreIndex
    index = VectorStoreIndex.from_documents(documents, storage_context=storage_context)

    # Query the index
    query_engine = index.as_query_engine()
    response = query_engine.query("What is this document about?")
    print(response.response)
else:
    print("No documents loaded. Skipping index creation and query.")

view raw JSON →