[Langgraph] Implementing a ReAct agent

Summary: Implementing a ReAct Agent with Langgraph

---

Overview

ReAct is a methodology that enables LLMs to interact with their environment (such as the web) to overcome hallucinations and error propagation issues that occur in traditional CoT (Chain of Thought) reasoning. Inspired by this paper, LangGraph has incorporated a prebuilt ReAct functionality. However, unlike the original meaning in the paper, this implementation refers to a design pattern where predefined tools are integrated into the LLM. This is further explained here.

---

Implementing a ReAct Agent

A tool is defined using input arguments and a docstring, which guide the LLM on how to utilize the tool effectively. The input arguments must be of type string, and the docstring is mandatory.

Example:

from typing import Literal
from langchain_core.tools import tool

@tool
def get_weather(city: Literal["nyc", "sf"]):
    """Use this to get weather information."""
    if city == "nyc":
        return "It might be cloudy in nyc"
    elif city == "sf":
        return "It's always sunny in sf"
    else:
        raise AssertionError("Unknown city")

This tool receives nyc or sf as input from the LLM and returns the corresponding weather information. The input values are constrained using Literal, and the docstring helps the LLM determine the appropriate input. (The internal mechanics of how this works are unclear.)

Another example:

from typing import Annotated, List
from langchain_core.tools import tool

@tool
def multiply_by_max(
    a: Annotated[int, "scale factor"],
    b: Annotated[List[int], "list of ints over which to take maximum"],
) -> int:
    """Multiply a by the maximum of b."""
    return a * max(b)

This tool takes a and b as inputs from the LLM and returns the product of a and the maximum value in b. The Annotated type adds descriptions to the input arguments.

Additionally, the schema of the tool’s arguments can be inspected using the following code:

print(multiply_by_max.args_schema.model_json_schema())

Now, let’s create an agent that can either provide weather information or perform mathematical calculations based on user input. This can be easily implemented using langgraph.prebuilt.create_react_agent.

from typing import Literal, Annotated, List
from langchain_core.tools import tool
from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

load_dotenv()  # Load API keys from the .env file

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

@tool
def get_weather(city: Literal["nyc", "sf"]):
    """Use this to get weather information."""
    if city == "nyc":
        return "It might be cloudy in nyc"
    elif city == "sf":
        return "It's always sunny in sf"
    else:
        raise AssertionError("Unknown city")

@tool
def multiply_by_max(
    a: Annotated[int, "scale factor"],
    b: Annotated[List[int], "list of ints over which to take maximum"],
) -> int:
    """Multiply a by the maximum of b."""
    return a * max(b)

tools = [get_weather, multiply_by_max]

graph = create_react_agent(model, tools=tools)

---

The Graph Structure

The graph can be visualized using the following code:

from IPython.display import Image, display

display(Image(graph.get_graph(xray=1).draw_mermaid_png()))

Now, let’s test the agent to see how the tools operate.

---

Using the ReAct Agent

from langchain_core.messages import HumanMessage

# Function to display streamed output in a clean format
def print_stream(stream):
    for s in stream:
        message = s["messages"][-1]
        if isinstance(message, tuple):
            print(message)
        else:
            message.pretty_print()


msg = {"messages": HumanMessage("What's the weather in nyc?")}  # User input
print_stream(graph.stream(msg, stream_mode="values"))  # Invoke the graph

Output:

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

What's the weather in nyc?
================================== Ai Message ==================================
Tool Calls:
get_weather (call_f72kfLYwXZhyNdiimAGjuYG4)
Call ID: call_f72kfLYwXZhyNdiimAGjuYG4
Args:
city: nyc
================================= Tool Message =================================
Name: get_weather

It might be cloudy in nyc
================================== Ai Message ==================================

The weather in NYC might be cloudy.

Now, let’s test the mathematical computation tool.

msg = {"messages": HumanMessage("Multiply 3 by the maximum of 1, 2, 9, 4, 5")}
print_stream(graph.stream(msg, stream_mode="values"))

Output:

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

Multiply 3 by the maximum of 1, 2, 9, 4, 5
================================== Ai Message ==================================
Tool Calls:
  multiply_by_max (call_9LniwXXq3XDPhdN1ALliL4hJ)
 Call ID: call_9LniwXXq3XDPhdN1ALliL4hJ
  Args:
    a: 3
    b: [1, 2, 9, 4, 5]
================================= Tool Message =================================
Name: multiply_by_max

27
================================== Ai Message ==================================

The result of multiplying 3 by the maximum of 1, 2, 9, 4, and 5 is 27.

---

Conclusion

Using Langgraph and Langchain’s tool system, implementing a ReAct agent is remarkably simple and efficient.