Tool Calling

What is Tool Calling?

• Allows LLMs to interact with other systems

• Supported by most of the newest LLMs, but not all

• Sounds complicated? Scary? It’s not too bad, actually…

Reference: https://jcheng5.github.io/llm-quickstart/quickstart.html#/how-it-works

How it does NOT work

How it DOES work

A tool is a function

• A tool call is a function that the LLM can use
• It can either infer what the function does by the function name, docstring, and/or parameter names
• You can also provide it the context you want

Example: Weather Tool

To ask the LLM about the weather in the current location we need to write a function that does a few things:

1. Geocode a location to a latitude and longitude (this can also be an API)
2. Use the latitude and longitude in an API that can look up the weather

Example Weather tool - Geocode

Example Weather tool - Geocode

{'lat': 40.7127281, 'lon': -74.0060152}

Example Weather tool - Weather

Example Weather tool - Weather

{'latitude': 40.710335,
 'longitude': -73.99309,
 'generationtime_ms': 0.06771087646484375,
 'utc_offset_seconds': 0,
 'timezone': 'GMT',
 'timezone_abbreviation': 'GMT',
 'elevation': 32.0,
 'current_weather_units': {'time': 'iso8601',
  'interval': 'seconds',
  'temperature': '°C',
  'windspeed': 'km/h',
  'winddirection': '°',
  'is_day': '',
  'weathercode': 'wmo code'},
 'current_weather': {'time': '2025-08-26T17:30',
  'interval': 900,
  'temperature': 25.5,
  'windspeed': 14.4,
  'winddirection': 193,
  'is_day': 1,
  'weathercode': 0}}

Example Weather tool - Register

from chatlas import ChatAnthropic

chat = ChatAnthropic()

chat.register_tool(get_coordinates)
chat.register_tool(get_weather)

Demo: Weather R

library(httr)
library(ellmer)
library(dotenv)

# Load environment variables
load_dot_env()

# Define weather function
get_weather <- function(latitude, longitude) {
  base_url <- "https://api.open-meteo.com/v1/forecast"

  tryCatch(
    {
      response <- GET(
        base_url,
        query = list(
          latitude = latitude,
          longitude = longitude,
          current = "temperature_2m,wind_speed_10m,relative_humidity_2m"
        )
      )
      rawToChar(response$content)
    },
    error = function(e) {
      paste("Error fetching weather data:", e$message)
    }
  )
}

# Create chat instance
chat <- chat_openai(
  model = "gpt-4.1",
  system_prompt = "You are a helpful assistant that can check the weather. Report results in imperial units."
)

# Register the weather tool
#
# Created using `ellmer::create_tool_def(get_weather)`
chat$register_tool(tool(
  get_weather,
  "Fetches weather information for a specified location given by latitude and
longitude.",
  latitude = type_number(
    "The latitude of the location for which weather information is requested."
  ),
  longitude = type_number(
    "The longitude of the location for which weather information is requested."
  )
))

# Test the chat
chat$chat("What is the weather in Seattle?")

Demo: Weather Python

import requests
from chatlas import ChatAnthropic
from dotenv import load_dotenv

load_dotenv()  # Loads OPENAI_API_KEY from the .env file


# Define a simple tool for getting the current weather
def get_weather(latitude: float, longitude: float):
    """
    Get the current weather for a location using latitude and longitude.
    """
    base_url = "https://api.open-meteo.com/v1/forecast"
    params = {
        "latitude": latitude,
        "longitude": longitude,
        "current": "temperature_2m,wind_speed_10m,relative_humidity_2m",
    }

    try:
        response = requests.get(base_url, params=params)
        response.raise_for_status()  # Raise an exception for bad status codes
        return response.text
    except requests.RequestException as e:
        return f"Error fetching weather data: {str(e)}"


chat = ChatAnthropic(
    model="claude-3-5-sonnet-latest",
    system_prompt=(
        "You are a helpful assistant that can check the weather. "
        "Report results in imperial units."
    ),
)

chat.register_tool(get_weather)
chat.chat("What is the weather in Seattle?")

Demo: Shiny Application

from chatlas import ChatAnthropic
from shiny.express import ui

from helper.get_coordinates import get_coordinates
from helper.get_weather import get_weather

chat_client = ChatAnthropic()

chat_client.register_tool(get_coordinates)
chat_client.register_tool(get_weather)

chat = ui.Chat(id="chat")
chat.ui(
    messages=[
        "Hello! I am a weather bot! Where would you like to get the weather form?"
    ]
)


@chat.on_user_submit
async def _(user_input: str):
    response = await chat_client.stream_async(user_input, content="all")
    await chat.append_message_stream(response)

You try: register a tool

• Functions: https://github.com/chendaniely/nydsaic2025-llm/tree/main/helper
• App: https://github.com/chendaniely/nydsaic2025-llm/blob/main/code/03-tools/app.py

Asking for permission (CLI only)

• When you want to explicitly pause and have user input to make a tool call.

from chatlas import Tool RejectError

def get_weather_check(lat: float, lon: float):
    allow = input_approval(
        "Would you like me to run the weather check function?"
    )
    if allow:
        return get_weather(lat, lon)
    raise ToolRejectError("The user has chosen to disallow the tool call.")