Function Calling

The protocol by which an LLM emits a structured function invocation that a runtime then executes — synonymous with tool calling.

Function calling was introduced by OpenAI in June 2023 as a way for GPT models to emit structured JSON describing a function to invoke. Anthropic shipped equivalent tool use in late 2023. The two are functionally similar and have converged in practice.

The protocol: the developer declares available functions with names, descriptions, and JSON schemas. The model, when it decides a function is needed, returns the function name plus arguments instead of (or alongside) natural-language text. The runtime executes the function and returns the result to the model. Without this primitive, agents would have to parse model output for tool intents — fragile and error-prone.