""" Function Calling and Tool Use Formatting ======================================== This example demonstrates how to format function definitions for LLM function calling, parse function calls from LLM outputs, and handle tool use patterns. Main concepts: - Formatting function definitions for LLMs - Creating tool definitions (OpenAI format) - Parsing function calls from LLM responses - Formatting function results for LLM consumption """ from kerb.parsing import ( format_function_call, format_tool_call, parse_function_call, format_function_result, ParseMode ) # Example functions that an LLM might call def get_weather(location: str, units: str = "celsius") -> dict: """Get weather information for a location.""" # Simulated weather data weather_data = { "New York": {"temp": 22, "condition": "Sunny", "humidity": 65}, "London": {"temp": 15, "condition": "Cloudy", "humidity": 80}, "Tokyo": {"temp": 28, "condition": "Clear", "humidity": 55} } return weather_data.get(location, {"temp": 20, "condition": "Unknown", "humidity": 50}) def search_database(query: str, limit: int = 10, filters: dict = None) -> list: """Search database with optional filters.""" # %% # Setup and Imports # ----------------- # Simulated search results return [ {"id": 1, "title": f"Result for {query}", "score": 0.95}, {"id": 2, "title": f"Another match for {query}", "score": 0.87} ] # %% # Send Email # ---------- def send_email(to: str, subject: str, body: str) -> dict: """Send an email.""" return {"status": "sent", "message_id": "msg_12345", "to": to} def simulate_llm_with_function_call(prompt: str) -> str: """Simulate LLM responses with function calls.""" if "weather" in prompt.lower(): return """{ "function": "get_weather", "arguments": { "location": "New York", "units": "celsius" } }""" elif "search" in prompt.lower(): return """I'll search the database for you. ```json { "function": "search_database", "arguments": { "query": "machine learning tutorials", "limit": 5, "filters": {"category": "education"} } } ```""" elif "email" in prompt.lower(): return """{ "function": "send_email", "arguments": { "to": "team@example.com", "subject": "Project Update", "body": "The project is on track for delivery next week." } }""" return "{}" # %% # Main # ---- def main(): """Run function calling examples.""" print("="*80) print("FUNCTION CALLING AND TOOL USE FORMATTING") print("="*80) # Example 1: Format function definition print("\nExample 1: Format Function Definition") print("-"*80) weather_func = format_function_call( name="get_weather", description="Get current weather information for a specific location", parameters={ "location": { "type": "string", "description": "City name (e.g., 'New York', 'London')" }, "units": { "type": "string", "enum": ["celsius", "fahrenheit"], "description": "Temperature units", "default": "celsius" } }, required=["location"] ) print("Weather Function Definition:") print(f" Name: {weather_func['name']}") print(f" Description: {weather_func['description']}") print(f" Required params: {weather_func['parameters'].get('required', [])}") print(f" Parameters: {', '.join(weather_func['parameters']['properties'].keys())}") # Example 2: Format tool definition (OpenAI format) print("\n\nExample 2: Format Tool Definition (OpenAI Format)") print("-"*80) search_tool = format_tool_call( name="search_database", description="Search the knowledge database with optional filters", parameters={ "query": { "type": "string", "description": "Search query string" }, "limit": { "type": "integer", "description": "Maximum number of results", "default": 10 }, "filters": { "type": "object", "description": "Optional filters to apply", "properties": { "category": {"type": "string"}, "date_range": {"type": "string"} } } }, required=["query"] ) print("Search Tool Definition:") print(f" Type: {search_tool['type']}") print(f" Function Name: {search_tool['function']['name']}") print(f" Description: {search_tool['function']['description']}") # Example 3: Parse function call from LLM output print("\n\nExample 3: Parse Function Call from LLM") print("-"*80) llm_output = simulate_llm_with_function_call("What's the weather in New York?") print(f"LLM Output:\n{llm_output}\n") result = parse_function_call(llm_output) if result.success: call_data = result.data print(f"Parsed Function Call:") print(f" Function: {call_data.get('function')}") print(f" Arguments: {call_data.get('arguments')}") # Execute the function func_name = call_data.get('function') args = call_data.get('arguments', {}) if func_name == "get_weather": weather = get_weather(**args) print(f"\nExecution Result: {weather}") # Example 4: Parse function call with markdown print("\n\nExample 4: Parse Function Call with Markdown") print("-"*80) llm_output = simulate_llm_with_function_call("search for tutorials") print(f"LLM Output:\n{llm_output}\n") result = parse_function_call(llm_output, mode=ParseMode.LENIENT) if result.success: call_data = result.data print(f"Parsed Function Call:") print(f" Function: {call_data.get('function')}") print(f" Arguments:") for key, value in call_data.get('arguments', {}).items(): print(f" - {key}: {value}") # Example 5: Format function result for LLM print("\n\nExample 5: Format Function Result for LLM") print("-"*80) function_name = "get_weather" function_result = {"temp": 22, "condition": "Sunny", "humidity": 65} formatted_result = format_function_result( result=function_result, name=function_name ) print(f"Function: {function_name}") print(f"Result: {function_result}\n") print(f"Formatted for LLM:\n{formatted_result}") # Example 6: Complete function calling workflow print("\n\nExample 6: Complete Function Calling Workflow") print("-"*80) # Step 1: User request user_request = "Send an email to the team about project status" print(f"User Request: {user_request}\n") # Step 2: LLM generates function call llm_output = simulate_llm_with_function_call(user_request) print(f"LLM Function Call:\n{llm_output}\n") # Step 3: Parse function call result = parse_function_call(llm_output) if result.success: call_data = result.data func_name = call_data.get('function') args = call_data.get('arguments', {}) print(f"Executing: {func_name}(**{args})\n") # Step 4: Execute function if func_name == "send_email": exec_result = send_email(**args) print(f"Execution Result: {exec_result}\n") # Step 5: Format result for LLM formatted = format_function_result(exec_result, name=func_name) print(f"Formatted Result for LLM:\n{formatted}") # Example 7: Multiple function definitions print("\n\nExample 7: Multiple Function Definitions") print("-"*80) available_functions = [ format_function_call( name="get_weather", description="Get weather for a location", parameters={"location": {"type": "string"}}, required=["location"] ), format_function_call( name="search_database", description="Search knowledge base", parameters={"query": {"type": "string"}}, required=["query"] ), format_function_call( name="send_email", description="Send an email", parameters={ "to": {"type": "string"}, "subject": {"type": "string"}, "body": {"type": "string"} }, required=["to", "subject", "body"] ) ] print("Available Functions:") for func in available_functions: print(f" - {func['name']}: {func['description']}") print("\n" + "="*80) print("Example completed successfully!") print("="*80) if __name__ == "__main__": main()