LangChain provides modules for managing and optimizing the use of large language models (LLMs) in applications. Its core philosophy is to facilitate data-aware applications where the language model interacts with other data sources and its environment. This framework consists of several parts that simplify the entire application lifecycle:

  • Write your applications in LangChain/LangChain.js. Get started quickly by using Templates for reference.
  • Use LangSmith to inspect, test, and monitor your chains to constantly improve and deploy with confidence.
  • Turn any chain into an API with LangServe.

By integrating Pinecone with LangChain, you can add knowledge to LLMs via retrieval augmented eneration (RAG), greatly enhancing LLM ability for autonomous agents, chatbots, question-answering, and multi-agent systems.

​
Setup guide

This guide shows you how to integrate Pinecone, a high-performance vector database, with LangChain, a framework for building applications powered by large language models (LLMs).

Pinecone enables developers to build scalable, real-time recommendation and search systems based on vector similarity search. LangChain, on the other hand, provides modules for managing and optimizing the use of language models in applications. Its core philosophy is to facilitate data-aware applications where the language model interacts with other data sources and its environment.

By integrating Pinecone with LangChain, you can add knowledge to LLMs via Retrieval Augmented Generation (RAG), greatly enhancing LLM ability for autonomous agents, chatbots, question-answering, and multi-agent systems.

This guide demonstrates only one way out of many that you can use LangChain and Pinecone together. For additional examples, see:

​
Key concepts

The PineconeVectorStore class provided by LangChain can be used to interact with Pinecone indexes. It’s important to remember that you must have an existing Pinecone index before you can create a PineconeVectorStore object.

​
Initializing a vector store

To initialize a PineconeVectorStore object, you must provide the name of the Pinecone index and an Embeddings object initialized through LangChain. There are two general approaches to initializing a PineconeVectorStore object:

  1. Initialize without adding records:
Python
    import os
    from langchain_pinecone import PineconeVectorStore
    from langchain_openai import OpenAIEmbeddings

    os.environ['OPENAI_API_KEY'] = '<YOUR_OPENAI_API_KEY>'
    os.environ['PINECONE_API_KEY'] = '<YOUR_PINECONE_API_KEY>'

    index_name = "<YOUR_PINECONE_INDEX_NAME>"
    embeddings = OpenAIEmbeddings()

    vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)
  1. Initialize while adding records:

The from_documents and from_texts methods of LangChain’s PineconeVectorStore class add records to a Pinecone index and return a PineconeVectorStore object.

The from_documents method accepts a list of LangChain’s Document class objects, which can be created using LangChain’s CharacterTextSplitter class. The from_texts method accepts a list of strings. Similarly to above, you must provide the name of an existing Pinecone index and an Embeddings object.

Both of these methods handle the embedding of the provided text data and the creation of records in your Pinecone index.

Python
    import os
    from langchain_pinecone import PineconeVectorStore
    from langchain_openai import OpenAIEmbeddings
    from langchain_community.document_loaders import TextLoader
    from langchain_text_splitters import CharacterTextSplitter

    os.environ['OPENAI_API_KEY'] = '<YOUR_OPENAI_API_KEY>'
    os.environ['PINECONE_API_KEY'] = '<YOUR_PINECONE_API_KEY>'

    index_name = "<YOUR_PINECONE_INDEX_NAME>"
    embeddings = OpenAIEmbeddings()

    # path to an example text file
    loader = TextLoader("../../modules/state_of_the_union.txt")
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    docs = text_splitter.split_documents(documents)

    vectorstore_from_docs = PineconeVectorStore.from_documents(
        docs,
        index_name=index_name,
        embedding=embeddings
    )

    texts = ["Tonight, I call on the Senate to: Pass the Freedom to Vote Act.", "ne of the most serious constitutional responsibilities a President has is nominating someone to serve on the United States Supreme Court.", "One of our nation’s top legal minds, who will continue Justice Breyer’s legacy of excellence."]

    vectorstore_from_texts = PineconeVectorStore.from_texts(
        texts,
        index_name=index_name,
        embedding=embeddings
    )

​
Add more records

Once you have initialized a PineconeVectorStore object, you can add more records to the underlying Pinecone index (and thus also the linked LangChain object) using either the add_documents or add_texts methods.

Like their counterparts that also initialize a PineconeVectorStore object, both of these methods also handle the embedding of the provided text data and the creation of records in your Pinecone index.

Python
    # path to an example text file
    loader = TextLoader("../../modules/inaugural_address.txt")
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    docs = text_splitter.split_documents(documents)

    vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)

    vectorstore.add_documents(docs)
Python
    vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)

    vectorstore.add_texts(["More text to embed and add to the index!"])

A similarity_search on a PineconeVectorStore object returns a list of LangChain Document objects most similar to the query provided. While the similarity_search uses a Pinecone query to find the most similar results, this method includes additional steps and returns results of a different type.

The similarity_search method accepts raw text and automatically embeds it using the Embedding object provided when you initialized the PineconeVectorStore. You can also provide a k value to determine the number of LangChain Document objects to return. The default value is k=4.

Python
    query = "Who is Ketanji Brown Jackson?"
    vectorstore.similarity_search(query)
    
    # Response:
    # [
    #    Document(page_content='Ketanji Onyika Brown Jackson is an American lawyer and jurist who is an associate justice of the Supreme Court of the United...', metadata={'chunk': 0.0, 'source': 'https://en.wikipedia.org/wiki/Ketanji_Brown_Jackson', 'title': 'Ketanji Brown Jackson', 'wiki-id': '6573'}),  
    #    Document(page_content='Jackson was nominated to the Supreme Court by President Joe Biden on February 25, 2022, and confirmed by the U.S. Senate...', metadata={'chunk': 1.0, 'source': 'https://en.wikipedia.org/wiki/Ketanji_Brown_Jackson', 'title': 'Ketanji Brown Jackson', 'wiki-id': '6573'}),  
    #    Document(page_content='Jackson grew up in Miami and attended Miami Palmetto Senior High School. She distinguished herself as a champion debater...', metadata={'chunk': 3.0, 'source': 'https://en.wikipedia.org/wiki/Ketanji_Brown_Jackson', 'title': 'Ketanji Brown Jackson', 'wiki-id': '6573'}),
    #    Document(page_content='After high school, Jackson matriculated at Harvard University to study government, having applied despite her guidance...', metadata={'chunk': 5.0, 'source': 'https://en.wikipedia.org/wiki/Ketanji_Brown_Jackson', 'title': 'Ketanji Brown Jackson', 'wiki-id': '6573'})
    # ]

You can also optionally apply a metadata filter to your similarity search. The filtering query language is the same as for Pinecone queries, as detailed in Filtering with metadata.

Python
    query = "Tell me more about Ketanji Brown Jackson."
    vectorstore.similarity_search(query, filter={'source': 'https://en.wikipedia.org/wiki/Ketanji_Brown_Jackson'})

​
Namespaces

Several methods of the PineconeVectorStore class support using namespaces. You can also initialize your PineconeVectorStore object with a namespace to restrict all further operations to that space.

Python
    index_name = "<YOUR_PINECONE_INDEX_NAME>"
    embeddings = OpenAIEmbeddings()

    vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings, namespace="example-namespace")

If you initialize your PineconeVectorStore object without a namespace, you can specify the target namespace within the operation.

Python
    # path to an example text file
    loader = TextLoader("../../modules/congressional_address.txt")
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
    docs = text_splitter.split_documents(documents)

    vectorstore_from_docs = PineconeVectorStore.from_documents(
        docs,
        index_name=index_name,
        embedding=embeddings,
        namespace="example-namespace"
    )

    vectorstore_from_texts = PineconeVectorStore.from_texts(
        texts,
        index_name=index_name,
        embedding=embeddings,
        namespace="example-namespace"
    )

    vectorstore_from_docs.add_documents(docs, namespace="example-namespace")

    vectorstore_from_texts.add_texts(["More text!"], namespace="example-namespace")
Python
    query = "Who is Ketanji Brown Jackson?"
    vectorstore.similarity_search(query, namesapce="example-namespace")

​
Tutorial

​
1. Set up your environment

Before you begin, install some necessary libraries and set environment variables for your Pinecone and OpenAI API keys:

pip install -qU \
  pinecone-client==3.0.0 \
  pinecone-datasets==0.7.0 \
  langchain-pinecone==0.0.3 \
  langchain-openai==0.0.7 \
  langchain==0.1.9

# Set environment variables for API keys
export PINECONE_API_KEY=<your Pinecone API key available at app.pinecone.io>
export OPENAI_API_KEY=<your OpenAI API key, available at platform.openai.com/api-keys>
Python
import os
pinecone_api_key = os.environ.get('PINECONE_API_KEY')
openai_api_key = os.environ.get('OPENAI_API_KEY')

​
2. Build the knowledge base

  1. Load a sample Pinecone dataset into memory:

    Python
    import pinecone_datasets  
dataset = pinecone_datasets.load_dataset('wikipedia-simple-text-embedding-ada-002-100K')  
len(dataset)  

# Response:
# 100000

  2. Reduce the dataset and format it for upserting into Pinecone:

    Python
    # we drop sparse_values as they are not needed for this example  
dataset.documents.drop(['metadata'], axis=1, inplace=True)  
dataset.documents.rename(columns={'blob': 'metadata'}, inplace=True)  
# we will use rows of the dataset up to index 30_000  
dataset.documents.drop(dataset.documents.index[30_000:], inplace=True)

​
3. Index the data in Pinecone

  1. Decide whether to use a serverless or pod-based index.

    Serverless indexes are in public preview and are available only on AWS in the us-west-2, us-east-1, and eu-west-1 regions. Check current limits and restrictions and test thoroughly before using them in production.

    Python 
    use_serverless = True

  2. Initialize your client connection to Pinecone and create an index. This step uses the Pinecone API key you set as an environment variable earlier.

    Python
    from pinecone import Pinecone, ServerlessSpec, PodSpec  
import time  
# configure client  
pc = Pinecone(api_key=pinecone_api_key)  
if use_serverless:  
    spec = ServerlessSpec(cloud='aws', region='us-east-1')  
else:  
    # if not using a starter index, you should specify a pod_type too  
    spec = PodSpec()  
# check for and delete index if already exists  
index_name = 'langchain-retrieval-augmentation-fast'  
if index_name in pc.list_indexes().names():  
    pc.delete_index(index_name)  
# create a new index  
pc.create_index(  
    index_name,  
    dimension=1536,  # dimensionality of text-embedding-ada-002  
    metric='dotproduct',  
    spec=spec  
)  
# wait for index to be initialized  
while not pc.describe_index(index_name).status['ready']:  
    time.sleep(1)

  3. Target the index and check its current stats:

    Python
    index = pc.Index(index_name)  
index.describe_index_stats()  

# Response:
# {'dimension': 1536,  
# 'index_fullness': 0.0,  
# 'namespaces': {},  
# 'total_vector_count': 0}

    You’ll see that the index has a total_vector_count of 0, as you haven’t added any vectors yet.

  4. Now upsert the data to Pinecone:

    Python
    for batch in dataset.iter_documents(batch_size=100):  
    index.upsert(batch)

  5. Once the data is indexed, check the index stats once again:

    Python
    index.describe_index_stats()  

# Response:
# {'dimension': 1536,  
# 'index_fullness': 0.0,  
# 'namespaces': {},  
# 'total_vector_count': 70000}

​
4. Initialize a LangChain vector store

Now that you’ve built your Pinecone index, you need to initialize a LangChain vector store using the index. This step uses the OpenAI API key you set as an environment variable earlier. Note that OpenAI is a paid service and so running the remainder of this tutorial may incur some small cost.

  1. Initialize a LangChain embedding object:

    Python
    from langchain_openai import OpenAIEmbeddings  
# get openai api key from platform.openai.com  
model_name = 'text-embedding-ada-002'  
embeddings = OpenAIEmbeddings(  
    model=model_name,  
    openai_api_key=openai_api_key  
)

  2. Initialize the LangChain vector store:

    The text_field parameter sets the name of the metadata field that stores the raw text when you upsert records using a LangChain operation such as vectorstore.from_documents or vectorstore.add_texts. This metadata field is used as the page_content in the Document objects retrieved from query-like LangChain operations such as vectorstore.similarity_search. If you do not specify a value for text_field, it will default to "text".

    Python
    from langchain_pinecone import PineconeVectorStore  
text_field = "text"  
vectorstore = PineconeVectorStore(  
    index, embeddings, text_field  
)

  3. Now you can query the vector store directly using vectorstore.similarity_search:

    Python
    query = "who was Benito Mussolini?"  
vectorstore.similarity_search(  
    query,  # our search query  
    k=3  # return 3 most relevant docs  
)  

# Response:
# [Document(page_content='Benito Amilcare Andrea Mussolini KSMOM GCTE (29 July 1883 – 28 April 1945) was an Italian politician and journalist...', metadata={'chunk': 0.0, 'source': 'https://simple.wikipedia.org/wiki/Benito%20Mussolini', 'title': 'Benito Mussolini', 'wiki-id': '6754'}),  
# Document(page_content='Fascism as practiced by Mussolini\nMussolini\'s form of Fascism, "Italian Fascism"- unlike Nazism, the racist ideology...', metadata={'chunk': 1.0, 'source': 'https://simple.wikipedia.org/wiki/Benito%20Mussolini', 'title': 'Benito Mussolini', 'wiki-id': '6754'}),  
# Document(page_content='Veneto was made part of Italy in 1866 after a war with Austria. Italian soldiers won Latium in 1870. That was when...', metadata={'chunk': 5.0, 'source': 'https://simple.wikipedia.org/wiki/Italy', 'title': 'Italy', 'wiki-id': '363'})]

All of these sample results are good and relevant. But what else can you do with this? There are many tasks, one of the most interesting (and well supported by LangChain) is called “Generative Question-Answering” or GQA.

​
5. Use Pinecone and LangChain for RAG

In RAG, you take the query as a question that is to be answered by a LLM, but the LLM must answer the question based on the information it is seeing from the vectorstore.

  1. To do this, initialize a RetrievalQA object like so:

    Python
    from langchain_openai import ChatOpenAI  
from langchain.chains import RetrievalQA  
# completion llm  
llm = ChatOpenAI(  
    openai_api_key=OPENAI_API_KEY,  
    model_name='gpt-3.5-turbo',  
    temperature=0.0  
)  
qa = RetrievalQA.from_chain_type(  
    llm=llm,  
    chain_type="stuff",  
    retriever=vectorstore.as_retriever()  
)  
qa.run(query)  

# Response:
# Benito Mussolini was an Italian politician and journalist who served as the Prime Minister of Italy from 1922 until 1943. He was the leader of the National Fascist Party and played a significant role in the rise of fascism in Italy...

  2. You can also include the sources of information that the LLM is using to answer your question using a slightly different version of RetrievalQA called RetrievalQAWithSourcesChain:

    Python
    from langchain.chains import RetrievalQAWithSourcesChain  
qa_with_sources = RetrievalQAWithSourcesChain.from_chain_type(  
    llm=llm,  
    chain_type="stuff",  
    retriever=vectorstore.as_retriever()  
)  
qa_with_sources(query)

# Response:
# {'question': 'who was Benito Mussolini?',  
# 'answer': "Benito Mussolini was an Italian politician and journalist who served as the Prime Minister of Italy from 1922 until 1943. He was the leader of the National Fascist Party and played a significant role in the rise of fascism in Italy...",  
# 'sources': 'https://simple.wikipedia.org/wiki/Benito%20Mussolini'}

​
6. Clean up

When you no longer need the index, use the delete_index operation to delete it:

Python
pc.delete_index(index_name)

Was this page helpful?

HaystackLlamaIndex