How to Use LangGraph for Agents: Simple Guide

To use langgraph for agents, you create an agent by defining its tasks and tools using Agent and Tool classes, then run it with agent.run(). LangGraph helps you build AI agents that can perform complex workflows by connecting language models and tools in a graph structure.

Syntax

The basic syntax to use LangGraph for agents involves creating Tool objects that define specific actions, then combining them into an Agent. You run the agent with the run() method to get results.

• Tool: Defines a callable action with a name and function.
• Agent: Combines tools and manages the workflow.
• run(): Executes the agent with input and returns output.

python

from langgraph import Tool, Agent

# Define a tool with a name and a function
def greet(name: str) -> str:
    return f"Hello, {name}!"

greet_tool = Tool(name="Greet", func=greet)

# Create an agent with the tool
agent = Agent(tools=[greet_tool])

# Run the agent
result = agent.run("Alice")
print(result)

Output

Hello, Alice!

Example

This example shows how to create a LangGraph agent that uses two tools: one to greet a user and another to say goodbye. The agent runs both tools in sequence and returns their outputs.

python

from langgraph import Tool, Agent

# Define greeting tool
def greet(name: str) -> str:
    return f"Hello, {name}!"

# Define farewell tool
def farewell(name: str) -> str:
    return f"Goodbye, {name}!"

# Create tools
greet_tool = Tool(name="Greet", func=greet)
farewell_tool = Tool(name="Farewell", func=farewell)

# Create agent with both tools
agent = Agent(tools=[greet_tool, farewell_tool])

# Run agent with input
result = agent.run("Bob")
print(result)

Output

['Hello, Bob!', 'Goodbye, Bob!']

Common Pitfalls

Common mistakes when using LangGraph agents include:

• Not defining tools with the correct function signature (must accept input and return output).
• Forgetting to include all tools in the agent's tool list.
• Passing input that tools cannot handle, causing errors.
• Expecting run() to return a single string when it returns a list of outputs if multiple tools are used.

python

from langgraph import Tool, Agent

# Wrong: Tool function missing return
def bad_tool(name: str):
    print(f"Hi {name}")  # No return

# Right: Tool function returns string
def good_tool(name: str) -> str:
    return f"Hi {name}"

bad = Tool(name="Bad", func=bad_tool)
good = Tool(name="Good", func=good_tool)

agent = Agent(tools=[bad, good])

# Running this will cause issues because bad_tool returns None
result = agent.run("Eve")
print(result)  # Output will include None which may cause problems

Output

[None, 'Hi Eve']

Quick Reference

Concept	Description
Tool	Defines an action with a name and a function that takes input and returns output.
Agent	Combines multiple tools to create a workflow or agent behavior.
run(input)	Executes the agent with the given input and returns outputs from tools.
Function Signature	Tool functions must accept input and return output (usually strings).
Output	If multiple tools, run() returns a list of outputs in order of tools.

Key Takeaways

Create tools by defining functions that take input and return output, then wrap them with Tool.

Combine tools into an Agent and run it with agent.run(input) to get results.

Ensure tool functions always return output; missing returns cause errors.

agent.run() returns a list of outputs if multiple tools are used.

Pass input compatible with all tools to avoid runtime errors.