Creating tools for agents in LangChain - Visual Walkthrough

Concept Flow - Creating tools for agents

Define Tool Function

↓

Wrap Function as Tool

↓

Create Agent with Tools

↓

Agent Receives Input

↓

Agent Selects Tool

↓

Tool Executes Function

↓

Return Tool Output to Agent

↓

Agent Responds to User

This flow shows how you define a tool function, wrap it as a tool, create an agent with it, and how the agent uses the tool to answer user input.

Execution Sample

LangChain

from langchain.agents import Tool, initialize_agent
from langchain.llms import OpenAI

def add_numbers(a, b):
    return a + b

add_tool = Tool(name="Adder", func=add_numbers, description="Adds two numbers")

Defines a simple add function, wraps it as a LangChain tool named 'Adder'.

Execution Table

Step	Action	Input	Tool Selected	Function Called	Output
1	Agent receives input	"Add 3 and 5"	Adder	add_numbers(3, 5)	8
2	Agent returns output	N/A	N/A	N/A	"The sum is 8"
3	Agent receives input	"Add 10 and 20"	Adder	add_numbers(10, 20)	30
4	Agent returns output	N/A	N/A	N/A	"The sum is 30"
5	Agent receives input	"Multiply 2 and 3"	No matching tool	N/A	N/A
6	Agent returns output	N/A	N/A	N/A	"Sorry, I can't do that."

💡 Agent stops when no matching tool is found or after responding to user input.

Variable Tracker

Variable	Start	After Step 1	After Step 3	After Step 5	Final
input_text	None	"Add 3 and 5"	"Add 10 and 20"	"Multiply 2 and 3"	"Multiply 2 and 3"
tool_selected	None	Adder	Adder	None	None
function_output	None	8	30	None	None
agent_response	None	"The sum is 8"	"The sum is 30"	"Sorry, I can't do that."	"Sorry, I can't do that."

Key Moments - 2 Insights

Why does the agent say it can't do something when I ask to multiply?

Because the agent only has the 'Adder' tool registered, it cannot find a tool to handle multiplication. See execution_table row 5 where no matching tool is found.

How does the agent know which tool to use?

The agent matches the user input to tool descriptions or names. If the input mentions adding, it selects the 'Adder' tool as shown in execution_table rows 1 and 3.

Visual Quiz - 3 Questions

Look at the execution_table at step 3. What is the output of the tool function?

A 30

B 5

C 8

D None

💡 Hint

Check the 'Output' column in execution_table row 3.

At which step does the agent fail to find a matching tool?

A Step 4

B Step 5

C Step 2

D Step 6

💡 Hint

Look for 'No matching tool' in the 'Tool Selected' column.

If you add a multiplication tool, how would the agent respond to 'Multiply 2 and 3'?

A It would still say it can't do that.

B It would call the add tool by mistake.

C It would call the multiplication tool and return 6.

D It would crash.

💡 Hint

Think about how the agent selects tools based on input matching.

Concept Snapshot

Creating tools for agents:
- Define a Python function for the tool's task.
- Wrap it with langchain's Tool(name, func, description).
- Create an agent with these tools.
- Agent matches user input to tool descriptions.
- Calls the tool function and returns output.
- If no tool matches, agent responds it can't help.

Full Transcript

This visual execution shows how to create tools for agents using LangChain. First, you define a function that does a task, like adding numbers. Then you wrap it as a Tool with a name and description. Next, you create an agent that knows about this tool. When the agent gets user input, it tries to find a tool that matches the request. If it finds one, it calls the tool's function with the right inputs and returns the result. If no tool matches, it tells the user it can't help. The execution table traces inputs, tool selection, function calls, and outputs step-by-step. The variable tracker shows how inputs and outputs change over time. Key moments clarify why the agent might say it can't do something and how it picks tools. The quiz tests understanding of these steps.