Skip to main content
Function calling is how your AI Voice Agent executes tools to perform actions, not just respond with text.

​
What is Function Calling?

Function calling allows your agent to:
  1. Identify when a tool is needed - Analyze user request
  2. Select the right tool - Choose appropriate function
  3. Call the function - Execute with correct parameters
  4. Use the result - Incorporate result into response

​
How It Works

​
Basic Flow

User: "What's the weather in London?"
  ↓
Agent: Analyzes request → Identifies need for weather tool
  ↓
Agent: Calls get_weather(location="London")
  ↓
Tool: Returns "Weather in London: sunny, 22°C"
  ↓
Agent: "The weather in London is sunny, 22°C."

​
Automatic Tool Selection

The agent automatically:
  • Understands function signatures - From type hints and docstrings
  • Maps user intent to functions - Matches requests to available tools
  • Extracts parameters - Parses user input to function parameters
  • Handles errors - Manages tool execution errors

​
Function Requirements

​
Type Hints

Functions should have type hints: 
def get_weather(location: str) -> str:
    """Get weather for a location."""
    return f"Weather in {location}: sunny, 22°C"

​
Docstrings

Docstrings help the agent understand the function: 
def calculate(expression: str) -> str:
    """Calculate a mathematical expression.
    
    Args:
        expression: A string containing a mathematical expression
                   (e.g., "15 * 23 + 10")
    
    Returns:
        The result of the calculation as a string
    """
    try:
        result = eval(expression)
        return str(result)
    except Exception as e:
        return f"Error: {str(e)}"

​
Tool Execution

​
Single Tool Call

# User: "What's the weather in London?"
# Agent calls: get_weather(location="London")
# Result: "Weather in London: sunny, 22°C"
# Agent responds: "The weather in London is sunny, 22°C."

​
Multiple Tool Calls

# User: "What's the weather in London and convert 100 USD to EUR?"
# Agent calls:
#   1. get_weather(location="London")
#   2. convert_currency(amount=100, from_currency="USD", to_currency="EUR")
# Agent combines results into response

​
Tool Chaining

# User: "Get the weather in London and then create a calendar event for tomorrow"
# Agent calls:
#   1. get_weather(location="London")
#   2. add_event(title="Check weather", date="tomorrow", time="10:00")
# Agent provides complete response with both results

​
Best Practices

​
Clear Function Names

# ✅ Good
def get_weather(location: str) -> str:
    """Get weather for a location."""

# ❌ Bad
def w(loc: str) -> str:
    """Get weather."""

​
Descriptive Docstrings

# ✅ Good
def convert_currency(amount: float, from_currency: str, to_currency: str) -> str:
    """Convert currency from one type to another.
    
    Args:
        amount: The amount to convert
        from_currency: Source currency code (e.g., "USD")
        to_currency: Target currency code (e.g., "EUR")
    
    Returns:
        Converted amount as formatted string
    """

# ❌ Bad
def convert(a: float, f: str, t: str) -> str:
    """Convert currency."""

​
Error Handling

def get_weather(location: str) -> str:
    """Get weather for a location."""
    try:
        result = weather_api.get(location)
        return f"Weather in {location}: {result}"
    except Exception as e:
        return f"Sorry, I couldn't get weather for {location}: {str(e)}"

​
Next Steps