Definition: LLM capability to generate structured function calls that can be executed by external systems.
— Source: NERVICO, Product Development Consultancy
What is Function Calling
Function calling is the capability of an LLM to generate structured function calls that can be executed by external systems. Instead of simply generating text, the model produces JSON objects that describe which function to invoke and with what parameters, allowing an application to execute that action in the real world. This capability is what connects language models with APIs, databases, and external tools.
How It Works
The developer defines a set of available functions with their names, descriptions, and parameter schemas. When the user makes a request that requires an external action, the model analyzes the intent and generates a call to the appropriate function with the correct parameters in a structured format. The application receives this call, executes the actual function, and returns the result to the model so it can generate a final response to the user. This cycle can be chained: the model can invoke multiple functions in sequence to complete complex tasks.
Why It Matters
Function calling transforms LLMs from passive text generators into agents capable of taking action. Without this capability, AI would be limited to conversations with no real-world consequences. For businesses, function calling is the foundation of AI assistants that can query inventories, create tickets, send emails, update CRMs, and execute any operation exposed by an API. It is the bridge between the model’s intelligence and the ability to execute.
Practical Example
An AI-powered sales assistant receives the request “schedule a demo with Acme company for Thursday at 10.” The model generates three sequential function calls: first it searches for the Acme contact in the CRM, then it checks the team calendar for availability, and finally it creates the event and sends the invitation. All without the user interacting with any of these tools directly.