Skip to main content

Glean LangChain Integration

Glean's official LangChain integration enables you to build powerful AI agents that can search and reason over your organization's knowledge using Python and the LangChain framework.

langchain-glean

Official LangChain integration for Glean's search and chat capabilities

Installation​

pip install -U langchain-glean

Configuration​

API Tokens​

You'll need Glean API credentials, and specifically a user-scoped API token. API Tokens require the following scopes: chat, search. You should speak to your Glean administrator to provision these tokens.

Configure Environment Variables​

Configure your Glean credentials by setting the following environment variables:

export GLEAN_SUBDOMAIN="your-glean-subdomain"
export GLEAN_API_TOKEN="your-glean-api-token"
export GLEAN_ACT_AS="user@example.com"  # Optional: Email to act as when making requests

Usage Examples​

Using the Retriever​

The GleanSearchRetriever allows you to search and retrieve documents from Glean:

from langchain_glean.retrievers import GleanSearchRetriever

# Initialize the retriever (will use environment variables)
retriever = GleanSearchRetriever()

# Search for documents
documents = retriever.invoke("quarterly sales report")

# Process the results
for doc in documents:
    print(f"Title: {doc.metadata.get('title')}")
    print(f"URL: {doc.metadata.get('url')}")
    print(f"Content: {doc.page_content}")
    print("---")

Building an Agent with Tools​

The GleanSearchTool can be used in LangChain agents to search Glean:

from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_glean.retrievers import GleanSearchRetriever
from langchain_glean.tools import GleanSearchTool

# Initialize the retriever
retriever = GleanSearchRetriever()

# Create the tool
glean_tool = GleanSearchTool(
    retriever=retriever,
    name="glean_search",
    description="Search for information in your organization's content using Glean."
)

# Create an agent with the tool
llm = ChatOpenAI(model="gpt-4o")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant with access to Glean search."),
    ("user", "{input}")
])

agent = create_openai_tools_agent(llm, [glean_tool], prompt)
agent_executor = AgentExecutor(agent=agent, tools=[glean_tool])

# Run the agent
response = agent_executor.invoke({"input": "Find the latest quarterly report"})
print(response["output"])

RAG with LangChain Chains​

You can integrate the retriever with LangChain chains for more complex workflows:

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI
from langchain_glean.retrievers import GleanSearchRetriever

# Initialize the retriever
retriever = GleanSearchRetriever()

# Create a prompt template
prompt = ChatPromptTemplate.from_template(
    """Answer the question based only on the context provided.

Context: {context}

Question: {question}"""
)

# Initialize the language model
llm = ChatOpenAI(model="gpt-4o")

# Format documents function
def format_docs(docs):
    return "\n\n".join(doc.page_content for doc in docs)

# Create the chain
chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

# Run the chain
result = chain.invoke("What were our Q2 sales results?")
print(result)

Advanced Usage​

Search Parameters​

You can customize your search by passing additional parameters:

# Search with additional parameters
documents = retriever.invoke(
    "quarterly sales report",
    page_size=5,  # Number of results to return
    disable_spellcheck=True,  # Disable spellcheck
    max_snippet_size=200  # Maximum snippet size
)

Custom Retriever Configuration​

Configure the retriever with custom settings:

# Initialize with custom settings
retriever = GleanSearchRetriever(
    subdomain="your-subdomain",  # Override environment variable
    api_token="your-api-token",  # Override environment variable
    act_as="user@example.com",   # Override environment variable
    page_size=10,               # Default number of results
    max_snippet_size=300        # Default snippet size
)

For the complete API documentation and implementation details, visit the GitHub repository.