Skip to main content

LangGraph integration

Copy for LLM

LangGraph is a framework for constructing stateful, multi-agent applications with large language models (LLMs). Developers use it to build multi-step agent workflows that call tools, APIs, and databases. For more details, check out the LangGraph documentation.

Help keep this page up to date

This integration uses a third-party service. If you find outdated content, please submit an issue on GitHub.

LangGraph support comes from the same langchain-apify package as the LangChain integration. This page covers binding Apify tools to a LangGraph agent. See the LangChain page for the full tool reference, tool set selection, and non-agent uses such as document loading and retrieval.

Quick start

Install the packages:

pip install langgraph langchain-apify langchain-openai

Then give a model one Apify tool and let it answer from live web data:

import os

from langchain_apify import ApifyRAGWebBrowserTool
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

os.environ["APIFY_TOKEN"] = "Your Apify API token"
os.environ["OPENAI_API_KEY"] = "Your OpenAI API key"

agent = create_react_agent(ChatOpenAI(model="gpt-5.4-mini"), [ApifyRAGWebBrowserTool()])
result = agent.invoke({"messages": [("human", "Search the web and tell me what Apify is.")]})
print(result["messages"][-1].content)

The rest of this page builds on that: several tools with streamed steps, a whole tool set at once, and any other Actor.

Use Apify with LangGraph

This guide shows how to use Apify Actors with LangGraph by building a ReAct agent that searches the web for TikTok profiles and extracts data from them, using two dedicated Apify tools: ApifyRAGWebBrowserTool, which wraps RAG Web Browser, for the search and ApifyTikTokScraperTool, which wraps TikTok Scraper, for the profile data.

Prerequisites

  • Apify API token: To use Apify Actors in LangGraph, you need an Apify API token. If you don't have one, you can learn how to obtain it in the Apify documentation.

  • OpenAI API key: To work with agents in LangGraph, you need an OpenAI API key. If you don't have one, you can get it from the OpenAI platform.

  • Python packages: You need to install the following Python packages:

    pip install langgraph langchain-apify langchain-openai

Build the TikTok profile search and analysis agent

First, import all required packages:

import os

from langchain_apify import ApifyRAGWebBrowserTool, ApifyTikTokScraperTool
from langchain_core.messages import HumanMessage
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

Next, set the environment variables for the Apify API token and OpenAI API key:

os.environ["OPENAI_API_KEY"] = "Your OpenAI API key"
os.environ["APIFY_TOKEN"] = "Your Apify API token"

Instantiate the LLM and the Apify tools:

llm = ChatOpenAI(model="gpt-5.4-mini")

browser = ApifyRAGWebBrowserTool()
tiktok = ApifyTikTokScraperTool()

Each tool wraps one Actor behind a simplified input schema, so the model calls it without knowing Actor IDs or Actor input schemas.

Register only the tools you need

The langchain-apify package ships 19 tools grouped into three sets. Every tool you register widens the model's decision space, which can cause wrong tool selection, slower responses, and higher token usage. See choosing the right tool set for the full list and the tool set imports.

Create the ReAct agent with the LLM and Apify tools:

tools = [browser, tiktok]
agent_executor = create_react_agent(llm, tools)

Finally, run the agent and stream the messages:

for state in agent_executor.stream(
    stream_mode="values",
    input={
        "messages": [
            HumanMessage(content="Search the web for OpenAI TikTok profile and analyze their profile.")
        ]
    }):
    state["messages"][-1].pretty_print()
Search and analysis may take some time

Each tool call runs a real Actor on the Apify platform, so the agent may take from seconds to minutes to finish.

You will see the agent's messages in the console, which will show each step of the agent's workflow. The output below is abbreviated:

================================ Human Message =================================

Search the web for OpenAI TikTok profile and analyze their profile.
================================== AI Message ==================================
Tool Calls:
  apify_rag_web_browser (call_y2rbmQ6gYJYC2lHzWJAoKDaq)
 Call ID: call_y2rbmQ6gYJYC2lHzWJAoKDaq
  Args:
    query: OpenAI TikTok profile
    max_results: 1

...

================================== AI Message ==================================
Tool Calls:
  apify_tiktok_scraper (call_yQ0mLqXvRp8bT3nZKcWuHsAe)
 Call ID: call_yQ0mLqXvRp8bT3nZKcWuHsAe
  Args:
    search_query: openai
    search_type: user
    max_results: 5

...

================================== AI Message ==================================

The OpenAI TikTok profile is "OpenAI (@openai) Official". Here are some key details
about the profile:

- **Description**: The profile features "low key research previews" and includes
  videos that showcase their various projects and research developments.
- **Content focus**: The posts primarily involve previews of OpenAI's research and
  various AI-related innovations.

...

If you want to test the whole example, you can simply create a new file, langgraph_integration.py, and copy the whole code into it.

import os

from langchain_apify import ApifyRAGWebBrowserTool, ApifyTikTokScraperTool
from langchain_core.messages import HumanMessage
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

os.environ["OPENAI_API_KEY"] = "Your OpenAI API key"
os.environ["APIFY_TOKEN"] = "Your Apify API token"

llm = ChatOpenAI(model="gpt-5.4-mini")

browser = ApifyRAGWebBrowserTool()
tiktok = ApifyTikTokScraperTool()

tools = [browser, tiktok]
agent_executor = create_react_agent(llm, tools)

for state in agent_executor.stream(
    stream_mode="values",
    input={
        "messages": [
            HumanMessage(content="Search the web for OpenAI TikTok profile and analyze their profile.")
        ]
    }):
    state["messages"][-1].pretty_print()

Bind a whole tool set

Instead of importing tools one by one, you can give the agent an entire category. Each list holds tool classes, so instantiate them before passing them to the agent:

from langchain_apify import APIFY_SEARCH_TOOLS, APIFY_SOCIAL_TOOLS
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

llm = ChatOpenAI(model="gpt-5.4-mini")
tools = [tool_cls() for tool_cls in APIFY_SEARCH_TOOLS + APIFY_SOCIAL_TOOLS]
agent_executor = create_react_agent(llm, tools)

Run any other Actor

Actors without a dedicated tool go through ApifyActorsTool, which binds to an agent the same way:

from langchain_apify import ApifyActorsTool
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

llm = ChatOpenAI(model="gpt-5.4-mini")
trends = ApifyActorsTool("apify/google-trends-scraper")
agent_executor = create_react_agent(llm, [trends])

Resources