AgentExecutor setup and configuration in LangChain - Mini Project: Build & Apply

AgentExecutor Setup and Configuration with Langchain

📖 Scenario: You are building a simple assistant that can run tasks using Langchain's AgentExecutor. This assistant will use a basic tool and a language model to respond to user input.

🎯 Goal: Create a Python script that sets up a Langchain AgentExecutor with a simple tool and a language model, then configures it to run a task.

📋 What You'll Learn

Create a list called tools with one tool named EchoTool that returns the input text.

Create a variable called llm that is an instance of OpenAI with temperature=0.

Create an AgentExecutor called agent_executor using initialize_agent with tools, llm, and agent='zero-shot-react-description'.

Configure agent_executor to run the input 'Say hello'.

💡 Why This Matters

🌍 Real World

AgentExecutor lets you build smart assistants that can use tools and language models to answer questions or perform tasks automatically.

💼 Career

Understanding how to set up and configure AgentExecutor is useful for roles in AI development, chatbot creation, and automation engineering.

1

Create the tools list with EchoTool

Create a list called tools containing one tool named EchoTool that returns the input text unchanged. Use Tool from langchain.agents with name='EchoTool' and a func that returns its input.

LangChain

# Your code here to create tools list with EchoTool

Hint

Define a function echo_func that returns its input, then create a Tool with that function.

2

Create the language model instance

Create a variable called llm that is an instance of OpenAI with temperature=0. Import OpenAI from langchain.llms.

LangChain

from langchain.llms import OpenAI

# Your code here to create llm with temperature=0

Hint

Import OpenAI and create llm with temperature=0.

3

Initialize the AgentExecutor

Create a variable called agent_executor by calling initialize_agent with tools, llm, and agent='zero-shot-react-description'. Import initialize_agent from langchain.agents.

LangChain

from langchain.agents import initialize_agent

# Your code here to create agent_executor using initialize_agent

Hint

Use initialize_agent with the correct parameters to create agent_executor.

4

Run the AgentExecutor with input

Call agent_executor.run with the string 'Say hello' to execute the agent.

LangChain

# Your code here to run agent_executor with 'Say hello'

Hint

Use agent_executor.run with the exact string 'Say hello'.

Practice

1. What is the primary purpose of AgentExecutor in Langchain?

easy

A. To connect language models with tools to automate tasks

B. To train new language models from scratch

C. To store data in a database

D. To create user interfaces for chatbots

Solution

1. Step 1: Understand AgentExecutor role

   AgentExecutor acts as a bridge between language models and external tools to perform tasks automatically.

2. Step 2: Compare options with this role

   Only To connect language models with tools to automate tasks describes connecting models and tools to automate tasks, which matches AgentExecutor's purpose.

3. Final Answer:

   To connect language models with tools to automate tasks -> Option A

4. Quick Check:

   AgentExecutor = Connect models and tools [OK]

Hint: AgentExecutor links models and tools for automation [OK]

Common Mistakes:

• Confusing AgentExecutor with model training
• Thinking it manages databases
• Assuming it builds user interfaces

2. Which of the following is the correct way to initialize an agent with tools in Langchain?

easy

A. agent = initialize_agent(agent='zero-shot-react-description', tools, llm)

B. agent = initialize_agent(tools, llm, agent='zero-shot-react-description', verbose=True)

C. agent = initialize_agent(tools, llm, verbose=False, agent='react-zero-shot')

D. agent = initialize_agent(llm, tools, agent='zero-shot-react-description')

Solution

1. Step 1: Recall initialize_agent parameter order

   The correct order is llm first, then tools, followed by named parameters like agent type.

2. Step 2: Check each option's order and parameters

   agent = initialize_agent(llm, tools, agent='zero-shot-react-description') correctly uses llm, tools, agent type string. Others have wrong order or wrong agent name.

3. Final Answer:

   agent = initialize_agent(llm, tools, agent='zero-shot-react-description') -> Option D

4. Quick Check:

   initialize_agent(llm, tools, ...) correct order [OK]

Hint: Remember: llm first, then tools in initialize_agent [OK]

Common Mistakes:

• Swapping llm and tools arguments
• Using incorrect agent type strings
• Omitting agent type parameter

3. Given this code snippet, what will be printed?

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

llm = OpenAI(temperature=0)
tools = []
agent = initialize_agent(llm, tools, agent='zero-shot-react-description', verbose=False)
response = agent.run('What is the capital of France?')
print(response)

medium

A. The agent returns an empty string

B. Error: No tools available

C. Paris

D. The agent returns the question text

Solution

1. Step 1: Understand agent with empty tools

   Even with no tools, the agent uses the language model to answer questions directly.

2. Step 2: Analyze the question and model behavior

   The question is simple and factual; the OpenAI model with temperature=0 returns a deterministic answer "Paris".

3. Final Answer:

   Paris -> Option C

4. Quick Check:

   Agent with no tools uses LLM answer [OK]

Hint: Agent uses LLM answer if no tools provided [OK]

Common Mistakes:

• Assuming error if tools list is empty
• Expecting empty or repeated question output
• Confusing verbose with output content

4. Identify the error in this agent initialization code:

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

llm = OpenAI(temperature=0)
tools = [Tool(name='Search', func=search_function)]
agent = initialize_agent(llm, tools, agent='zero-shot-react-description', verbose=True)

medium

A. The Tool class is not imported

B. The order of arguments in initialize_agent is incorrect

C. temperature parameter is invalid for OpenAI

D. verbose parameter cannot be True

Solution

1. Step 1: Check imports for Tool usage

   The code uses Tool but does not import it from langchain.tools.

2. Step 2: Verify other parameters

   Argument order llm then tools is correct; temperature=0 is valid; verbose=True is allowed.

3. Final Answer:

   The Tool class is not imported -> Option A

4. Quick Check:

   import Tool from langchain.tools required [OK]

Hint: Import Tool from langchain.tools before using [OK]

Common Mistakes:

• Misidentifying argument order as error
• Overlooking missing Tool import
• Misunderstanding verbose usage

5. You want to create an AgentExecutor that uses two tools: a calculator and a search tool. Which setup correctly configures the agent to use both tools and logs detailed steps?

hard

A. tools = [CalculatorTool(), SearchTool()]
agent = initialize_agent(tools, llm, agent='zero-shot-react-description', verbose=True)

B. tools = [CalculatorTool(), SearchTool()]
agent = initialize_agent(llm, tools, agent='zero-shot-react-description', verbose=True)

C. tools = [CalculatorTool(), SearchTool()]
agent = initialize_agent(tools, llm, verbose=False, agent='zero-shot-react-description')

D. tools = [CalculatorTool(), SearchTool()]
agent = initialize_agent(tools, llm, agent='react-zero-shot', verbose=True)

Solution

1. Step 1: Confirm tools list and order

   Both CalculatorTool and SearchTool are included in a list assigned to tools, which is correct.

2. Step 2: Check initialize_agent parameters

   tools = [CalculatorTool(), SearchTool()] agent = initialize_agent(llm, tools, agent='zero-shot-react-description', verbose=True) uses correct order (llm, tools), correct agent type string, and verbose=True for detailed logs.

3. Final Answer:

   tools = [CalculatorTool(), SearchTool()] agent = initialize_agent(llm, tools, agent='zero-shot-react-description', verbose=True) -> Option B

4. Quick Check:

   Correct tools, order, agent type, and verbose [OK]

Hint: Use llm first then tools list, verbose=True for detailed logs [OK]

Common Mistakes:

• Swapping llm and tools arguments
• Using wrong agent type string
• Setting verbose to False when detailed logs needed