Creating an Agent With Custom Tools

AI Agents with Hugging Face smolagents

Adel Nehme

VP of AI Curriculum, DataCamp

Benefits of Custom Tools

Reliability: Write and test logic explicitly instead of relying on the agent to guess
Reusability: Use tools across projects and agents
Consistency: Get predictable behavior across runs (great for debugging)
Controlled access: Expose only what you want (files, APIs, databases, etc.)

Scenario: You Run a Retail Store

Inventory data is stored in a CSV file (size, color, quantity, and price)
Code agents can write code to read CSVs
But they don't have access to files by default
You need to wrap file access in a custom tool

Anatomy of a Custom Tool

from smolagents import tool
import pandas as pd

@tool
def check_inventory(product_name: str) -> int:
    """
    Check the available quantity of a product in the inventory CSV.

    Args:
        product_name (str): The name of the product to look up.

    Returns:
        int: The quantity in stock. Returns 0 if the product is not found.
    """
    df = pd.read_csv("store_inventory.csv")
    match = df[df["product_name"] == product_name]
    stock_quantity = int(match.iloc[0]["quantity"]) if not match.empty else 0
    return stock_quantity

Best Practices for Custom Tools

Do we have any t-shirts in stock?

Registering a Custom Tool with Your Agent

from smolagents import CodeAgent

agent = CodeAgent(
    tools=[check_inventory], # Add custom tool
    model=InferenceClientModel(),
    additional_authorized_imports=["pandas"]  # Allow external package
)

agent.run("Do we have any large t-shirts in stock?")

Yes, we have 8 large t-shirts in stock.

Custom Tools in Production Projects

Let's practice!

AI Agents with Hugging Face smolagents