Function Calling: Definition and Examples

Function Calling is a mechanism that allows language models to interact with external systems by generating structured function calls. Rather than simply producing text, the model can decide that a concrete action is necessary—such as querying a database, consulting a weather API, or performing a calculation—and produce a JSON object precisely describing the function to call and its parameters.

Concretely, the developer defines a set of available functions (with their name, description, and parameter schema) in the request sent to the model. When the user asks a question that requires one of these functions, the model does not respond directly: it returns a structured function call that the application can execute. The execution result is then returned to the model, which formulates a final response in natural language.

This approach solves a fundamental limitation of LLMs: their inability to access real-time data or perform actions in the real world. Thanks to Function Calling, a chatbot can book a flight, an assistant can check your calendar, or an agent can modify a file. The model acts as an intelligent router that understands the user's intention and translates it into concrete actions.

Function Calling is at the heart of the autonomous AI agent trend. It is the fundamental building block that transforms a simple text generator into a system capable of acting, observing results, and iterating. Major providers (OpenAI, Anthropic, Google) all offer this feature, each with its own implementation (tool use at Anthropic, function calling at OpenAI, etc.).