Using LangChain Tools: Add Web Search, Math, and Custom Actions

🔍 Give Your LLM Real Abilities

LLMs can reason. But they can't calculate, search the web, or hit APIs — unless you give them tools.

LangChain lets you attach tools to your LLMs so they can take actions, not just give answers.

This is the foundation for agents, copilots, and advanced LLM workflows.

🛠️ Common Tools You Can Add

• Web search (SerpAPI, Tavily, DuckDuckGo)
• Math calculator
• Python code executor
• API fetchers (custom tool wrappers)
• Database query runners

⚙️ Example: Add a Calculator Tool

from langchain.agents import initialize_agent, Tool
from langchain.llms import OpenAI
from langchain.agents.agent_types import AgentType
from langchain.tools import DuckDuckGoSearchRun
from langchain.tools.python.tool import PythonREPLTool

tools = [
    PythonREPLTool(),  # For math or Python logic
    DuckDuckGoSearchRun(name="Search")
]

llm = OpenAI()

agent = initialize_agent(
    tools,
    llm,
    agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
    verbose=True
)

agent.run("What is 17 * 23, and who won the 2023 World Cup?")

The agent will:

• Decide it needs to calculate 17*23
• Then use the search tool to answer the second part
• Return the combined response

💡 How Tools Work

LangChain agents follow a think → act → observe → repeat cycle.

• Think: Decide what to do
• Act: Call a tool
• Observe: See the result
• Repeat: Chain actions until task is complete

This architecture allows your LLM to use tools as functions, enabling complex workflows.

🔚 Wrapping Up This Section

By now, you've built:

• A basic Q&A app
• A chatbot with memory
• A tool-using LLM agent

In Chapter 5, we'll bring it all together with vector search and knowledge retrieval — powering apps like "Chat with Your Docs."