DEV Community

Vikas Choubey
Vikas Choubey

Posted on

🤖 Build a Weather Search Agent (LangChain + Ollama)

#python #agents #ai #developer

In this tutorial, we’ll build a simple AI agent that can fetch real-time weather information for a given country.

We’ll keep it:

  • 💻 Fully local (LLM via Ollama)
  • 🌐 Real-time (weather via API)
  • 🧠 Agent-based (tool usage with LangChain)

🎯 What You’ll Build

An agent that can answer:

“What’s the weather in India?”

Flow:


User → Agent → Tool (Weather API) → Agent → Response
Enter fullscreen mode Exit fullscreen mode

⚠️ Important Reality

Local models (via Ollama) cannot browse the internet.

So we solve this by:

👉 Giving the agent a tool that calls a real API

👉 The agent decides when to use that tool

🧱 Project Structure 


root/
│
├── main.py
└── projects/
└── search_agent/
├── app.py
└── tools.py
Enter fullscreen mode Exit fullscreen mode

🔑 Step 1: Get Free Weather API

Use: https://openweathermap.org/api

  1. Sign up
  2. Get API key

📦 Step 2: Install Dependencies 

pip install langchain langchain-ollama requests python-dotenv
Enter fullscreen mode Exit fullscreen mode

🔐 Step 3: Add ".env" 

WEATHER_API_KEY=your_api_key_here
Enter fullscreen mode Exit fullscreen mode

🛠 Step 4: Create Tool (REAL API)

📄 projects/search_agent/tools.py 

import requests
import os
from langchain.tools import tool


@tool
def get_weather(city: str) -> str:
    """
    Fetches real-time weather data for a given city.
    """
    api_key = os.getenv("WEATHER_API_KEY")

    url = f"http://api.openweathermap.org/data/2.5/weather?q={city}&appid={api_key}&units=metric"

    response = requests.get(url)

    if response.status_code != 200:
        return "Could not fetch weather data."

    data = response.json()

    weather = data["weather"][0]["description"]
    temp = data["main"]["temp"]

    return f"The current weather in {city} is {weather} with temperature {temp}°C."
Enter fullscreen mode Exit fullscreen mode

🧠 What is @tool?

The @tool decorator converts a Python function into a LangChain Tool.

Without @tool: 

def get_weather(city):
    ...
Enter fullscreen mode Exit fullscreen mode

Just a normal function.

With @tool: 

@tool
def get_weather(city):
    ...
Enter fullscreen mode Exit fullscreen mode

Now it becomes:

  • Discoverable by the agent
  • Has a name + description
  • Callable by LLM reasoning

Internally, it gives the agent:

  • Tool name: get_weather
  • Description (docstring)
  • Input schema

So the agent can decide:

Thought: I need weather info
Action: get_weather
Action Input: India
Enter fullscreen mode Exit fullscreen mode

⚙️ Step 5: Build the Agent

📄 projects/search_agent/app.py 

from langchain_ollama import ChatOllama
from langchain.agents import initialize_agent, AgentType
from .tools import get_weather


def search_agent():
    print("🌦 Weather Agent Started")

    llm = ChatOllama(
        model="llama3",
        temperature=0
    )

    tools = [get_weather]

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

    response = agent.run("What is the weather in India?")

    print("\nFinal Answer:")
    print(response)
Enter fullscreen mode Exit fullscreen mode

🧾 Step 6: Root Entry Point

📄 main.py 

from projects.search_agent.app import search_agent
from dotenv import load_dotenv

load_dotenv()


def main():
    print("Hello from LangChain Course!")
    search_agent()


if __name__ == "__main__":
    main()
Enter fullscreen mode Exit fullscreen mode

▶️ Step 7: Run the Project 

python main.py
Enter fullscreen mode Exit fullscreen mode

🔄 Agent Flow Diagram 

User Question
     ↓
LLM (Reasoning)
     ↓
Decides Tool → get_weather
     ↓
API Call (OpenWeather)
     ↓
Returns Data
     ↓
LLM formats answer
     ↓
Final Output
Enter fullscreen mode Exit fullscreen mode

🧠 What’s Happening Under the Hood

The agent follows ReAct pattern:

Thought → Action → Observation → Thought → Final Answer
Enter fullscreen mode Exit fullscreen mode

Example:

Thought: I need weather info
Action: get_weather
Action Input: India
Observation: The weather is 30°C, cloudy
Final Answer: The weather in India is...
Enter fullscreen mode Exit fullscreen mode

⚠️ Common Mistakes

❌ Expecting LLM to fetch real-time data
❌ Not using tools
❌ Missing API key
❌ Weak tool descriptions

🚀 Improvements You Can Add

  • 🌍 Country → City mapping
  • 🧠 Multi-step queries (forecast + humidity)
  • 💬 User input instead of hardcoded query
  • 📡 Add caching layer
  • 🧪 Add error handling + retries

🎯 Key Learnings

  • Agents don’t “know” real-time data
  • Tools bridge LLM → real world
  • @tool is the gateway to agent capabilities
  • ReAct loop powers decision making

🔥 What Next?

You can now build:

  • 🔎 Search agent (Google / Tavily)
  • 💻 Coding agent (file tools)
  • 📄 RAG agent (documents)
  • 🧠 Multi-agent system

You’ve officially moved from:

👉 LLM user → AI Agent builder 🚀

Top comments (1)

Collapse
 
klement_gunndu profile image
klement Gunndu •

Neat walkthrough — the @tool decorator explanation with the before/after comparison is the clearest I have seen. One thing worth adding: the agent tool-calling behavior changes significantly depending on which Ollama model you use. Llama3 handles tool schemas well but some smaller models silently skip the tool call entirely.