Skip to main content

LangChain

Build autonomous data agents with LangChain that are grounded in your enterprise data context from DataHub — ownership, lineage, documentation, quality signals, and more.

Prerequisites​

  • Python 3.10+
  • A DataHub instance and access token
  • An OpenAI API key (or another LLM provider with tool-calling support)

Installation​

pip install "datahub-agent-context[langchain]" langchain-openai

The [langchain] extra includes langchain and langchain-core. langchain-openai is a separate install for the OpenAI provider shown below.

Quick Start​

from datahub.sdk.main_client import DataHubClient
from datahub_agent_context.langchain_tools import build_langchain_tools

client = DataHubClient.from_env()

# Read-only tools by default; set include_mutations=True for write operations
tools = build_langchain_tools(client, include_mutations=False)

Wire the tools into a LangChain agent:

from langchain_openai import ChatOpenAI
from langchain.agents import create_agent

llm = ChatOpenAI(model="gpt-4o", temperature=0)  # any tool-calling model works here

agent = create_agent(
    llm,
    tools=tools,
    system_prompt="You are a data catalog assistant. Use the available tools to search for datasets, get entity details, and trace lineage. Always include URNs in your answers.",
)

response = agent.invoke({"messages": [{"role": "user", "content": "Find all datasets about customers"}]})
print(response["messages"][-1].content)

Prefer Bedrock? Swap only the model lines — tools, create_agent, and invoke stay identical:

import boto3
from langchain_aws import ChatBedrock  # pip install langchain-aws boto3

bedrock_runtime = boto3.client(service_name="bedrock-runtime", region_name="us-west-2")
llm = ChatBedrock(
    client=bedrock_runtime,
    model_id="us.anthropic.claude-haiku-4-5-20251001-v1:0",
    model_kwargs={"max_tokens": 2048, "temperature": 0},
)

Full working example: simple_search.py

Troubleshooting​

  • Tool execution errors? Verify your DataHub connection with client.test_connection() and check token permissions.
  • Agent not using tools? Strengthen the system prompt, or try temperature=0 for more deterministic tool use.
  • Import errors? Run pip install "datahub-agent-context[langchain]" langchain-openai.

Links: LangChain Agent Docs · Agent Context Kit · MCP Server Guide