langchain-glean 0.4.0

An integration package connecting Glean and LangChain

langchain-glean

Connect Glean – The Work AI platform connected to all your data – with LangChain.

Overview

The package provides:

• Chat models – ChatGlean wraps the /v1/chat assistant API and ChatGleanAgent targets specific agents
• Retrievers – typed helpers for Glean search and the people directory
• Tools – drop-in utilities for LangChain agents

Each implementation supports three input styles so you can start simple and scale up only when required:

1. Plain strings (retrievers only) – pass the search query text
2. Simple objects – pass a small Pydantic model (e.g. ChatBasicRequest, SearchBasicRequest, PeopleProfileBasicRequest)
3. Full Glean request classes – hand-craft a glean.models.SearchRequest, glean.models.ChatRequest, or glean.models.ListEntitiesRequest

Installation

pip install -U langchain-glean

Environment variables

export GLEAN_API_TOKEN="<your-token>"              # user or global token
export GLEAN_SERVER_URL="https://your-company-be.glean.com" # full backend URL (preferred)
# Deprecated: GLEAN_INSTANCE is still supported as a fallback
# export GLEAN_INSTANCE="acme"
export GLEAN_ACT_AS="[e-mail]"                # only for global tokens

Quick Start

Chat with Glean Assistant

from langchain_core.messages import HumanMessage
from langchain_glean.chat_models import ChatGlean

chat = ChatGlean()
response = chat.invoke([HumanMessage(content="When is the next company holiday?")])
print(response.content)

Need streaming? Replace invoke with stream or astream.

Chat Models

ChatGlean

Connect to the Glean Assistant API with LangChain's chat interface.

Basic usage with message list

from langchain_glean.chat_models import ChatGlean
from langchain_core.messages import HumanMessage

chat = ChatGlean()
response = chat.invoke([HumanMessage(content="Hello")])

Using the simplified request object

from langchain_glean.chat_models import ChatGlean, ChatBasicRequest

chat = ChatGlean()
response = chat.invoke(ChatBasicRequest(
    message="Hello", 
    context=["Previous conversation context"]
))

Using the full Glean SDK request

from glean.api_client import models
from langchain_glean.chat_models import ChatGlean

chat = ChatGlean()
req = models.ChatRequest(
    messages=[
        models.ChatMessage(
            author="USER", 
            message_type="CONTENT", 
            fragments=[models.ChatMessageFragment(text="Hello")]
        )
    ]
)
response = chat.invoke(req)

ChatGleanAgent

Connect directly to an individual Glean agent.

from langchain_core.messages import HumanMessage
from langchain_glean.chat_models import ChatGleanAgent

agent_chat = ChatGleanAgent(agent_id="abc123")
response = agent_chat.invoke([HumanMessage(content="What are our Q4 sales targets?")])
print(response.content)

You can provide additional input fields for the agent:

response = agent_chat.invoke(
    [HumanMessage(content="What are our sales targets?")],
    fields={"department": "Marketing", "quarter": "Q4"}
)

Retrievers

GleanSearchRetriever

Search Glean's unified index and get results as LangChain documents.

Simple string query

from langchain_glean.retrievers import GleanSearchRetriever

retriever = GleanSearchRetriever()
results = retriever.invoke("quarterly report")

Using the simplified request object

from langchain_glean.retrievers import GleanSearchRetriever, SearchBasicRequest

retriever = GleanSearchRetriever()
results = retriever.invoke(SearchBasicRequest(
    query="quarterly report", 
    data_sources=["confluence", "drive"]
))

Using the full Glean SDK request

from glean.api_client import models
from langchain_glean.retrievers import GleanSearchRetriever

retriever = GleanSearchRetriever()
req = models.SearchRequest(
    query="quarterly report", 
    page_size=5,
    facet_filters=[models.FacetFilter(name="datasource", values=["confluence"])]
)
results = retriever.invoke(req)

GleanPeopleProfileRetriever

Search Glean's people directory and get results as LangChain documents.

Simple string query

from langchain_glean.retrievers import GleanPeopleProfileRetriever

people = GleanPeopleProfileRetriever()
results = people.invoke("jane doe")

Using the simplified request object

from langchain_glean.retrievers import GleanPeopleProfileRetriever, PeopleProfileBasicRequest

people = GleanPeopleProfileRetriever()
results = people.invoke(PeopleProfileBasicRequest(
    query="staff engineer", 
    page_size=3
))

Using the full Glean SDK request

from glean.api_client import models
from langchain_glean.retrievers import GleanPeopleProfileRetriever

people = GleanPeopleProfileRetriever()
req = models.ListEntitiesRequest(
    entity_type="PEOPLE", 
    query="staff engineer", 
    page_size=3
)
results = people.invoke(req)

Tools for LangChain Agents

Available Tools

Tool name	Purpose	Arguments
GleanSearchTool	Search content	Query string or SearchBasicRequest
GleanPeopleProfileSearchTool	Find people	Query string or PeopleProfileBasicRequest
GleanChatTool	Converse with Glean Assistant	Message string or ChatBasicRequest
GleanListAgentsTool	List available agents	None
GleanGetAgentSchemaTool	Get agent input schema	agent_id (str)
GleanRunAgentTool	Run a specific agent	agent_id (str), fields (dict)

Basic Search Tool Example

from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_core.prompts import ChatPromptTemplate

from langchain_glean.retrievers import GleanSearchRetriever
from langchain_glean.tools import GleanSearchTool

# Create the tool
retriever = GleanSearchRetriever()
search_tool = GleanSearchTool(retriever=retriever)

# Set up the agent
llm = ChatOpenAI(model="gpt-3.5-turbo")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You can search our knowledge base when needed."),
    ("user", "{input}"),
])

agent = create_openai_tools_agent(llm, [search_tool], prompt)
executor = AgentExecutor(agent=agent, tools=[search_tool])

# Run the agent
result = executor.invoke({"input": "Find the latest QBR deck"})
print(result["output"])

Using Glean Agent Tools

from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_core.prompts import ChatPromptTemplate
from langchain_glean.tools import GleanListAgentsTool, GleanGetAgentSchemaTool, GleanRunAgentTool

# Create tools
list_agents_tool = GleanListAgentsTool()
get_schema_tool = GleanGetAgentSchemaTool()
run_agent_tool = GleanRunAgentTool()

# Set up the agent
llm = ChatOpenAI(model="gpt-4")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You can find and run Glean agents to answer questions."),
    ("user", "{input}"),
])

agent = create_openai_tools_agent(llm, [list_agents_tool, get_schema_tool, run_agent_tool], prompt)
executor = AgentExecutor(agent=agent, tools=[list_agents_tool, get_schema_tool, run_agent_tool])

# Run the agent
result = executor.invoke({
    "input": "Find an agent that can help with sales data and run it to get Q4 forecast"
})
print(result["output"])

Advanced Usage

Full Request Objects

Pass any SDK request class (SearchRequest, ChatRequest, ListEntitiesRequest) directly for complete API control.

Async Support

Every retriever and tool exposes ainvoke and async streams:

# Async search
documents = await retriever.ainvoke("monthly revenue")

# Async streaming chat
async for chunk in chat.astream([HumanMessage(content="Hello")]):
    print(chunk.content, end="", flush=True)

Custom Agent Config

Override model behavior per call:

chat.invoke(
    ChatBasicRequest(message="Summarise last quarter"),
    agent_config={"agent": "GPT", "mode": "SEARCH"},
    timeout_millis=30_000,
)

Resume a Chat

Either pass chat_id per call or set it as a property:

chat = ChatGlean()
chat.chat_id = "abc123"
chat.invoke([HumanMessage(content="Continue...")])

Contributing

1. mise install && mise run setup
2. mise run lint && mise run test
3. Open a PR!

Links

• Glean API – https://developer.glean.com
• LangChain docs – https://python.langchain.com
• Source code – https://github.com/langchain-ai/langchain-glean