AI & LLMs

LangGraph: A Guide to Building Stateful AI Agents

Zachary Carciu •
Advertisement

LangGraph: A Guide to Building Stateful AI Agents

LangGraph is a powerful framework for building complex AI agent workflows using stateful graphs. It allows for structured, conditional, and parallel execution of tasks, making it ideal for sophisticated AI applications.

1. Setting Up LangGraph

Installation

To install LangGraph, run the following command:

pip install langgraph

Import Required Modules

from typing import Annotated
from typing_extensions import TypedDict
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages

2. Creating a Basic LangGraph Agent

Define the State

The state defines the agent’s conversation history:

class State(TypedDict):
    messages: Annotated[list, add_messages]

Build the Graph

Creating a stateful graph involves defining nodes and connecting them with edges:

graph_builder = StateGraph(State)

def chatbot(state: State):
    # Implement LLM logic here
    llm_response = "Hello! How can I assist you?"
    return {"messages": [llm_response]}

graph_builder.add_node("chatbot", chatbot)

graph_builder.add_edge(START, "chatbot")
graph_builder.add_edge("chatbot", END)

# Compile the graph
graph = graph_builder.compile()

3. Adding Tools and Conditional Logic

Define External Tools

You can integrate tools such as web search or API calls:

from langchain_community.tools.tavily_search import TavilySearchResults

tool = TavilySearchResults(max_results=2)
tools = [tool]

Create a Tool Node

from langgraph.prebuilt import ToolNode, tools_condition

tool_node = ToolNode(tools=[tool])
graph_builder.add_node("tools", tool_node)

Add Conditional Edges

Using conditional logic, the agent can decide whether to invoke a tool:

graph_builder.add_conditional_edges(
    "chatbot",
    tools_condition,
    {
        "tools": "tools",
        "end": END
    }
)
graph_builder.add_edge("tools", "chatbot")

4. Running the LangGraph Agent

To invoke the agent, provide a message input:

from langchain.schema import HumanMessage

messages = [HumanMessage(content="Your query here")]
result = graph.invoke({"messages": messages})
print(result['messages'][-1].content)

5. Advanced Features

1. State Persistence

LangGraph allows saving and restoring agent states, making it ideal for long-running tasks.

2. Graph Visualization

You can visualize the graph structure with:

graph.draw_ascii()

Or generate an image representation:

graph.draw_mermaid_png("graph.png")

3. Parallel Execution

LangGraph supports parallel execution of API calls, improving efficiency for complex workflows.

6. Summary

LangGraph provides a structured approach to AI workflows, offering stateful graphs, conditional execution, and parallel processing. It is a powerful tool for building robust AI agents.

References

With this guide, you can start building sophisticated AI workflows using LangGraph!

Advertisement