Function Calling

Function calling (also called tool use or tool calling) is an LLM capability that lets a model invoke predefined functions or APIs as part of generating its response. You describe the available functions to the model — name, purpose, and JSON-schema parameters — and the model decides during inference when calling one is appropriate. It returns structured JSON arguments instead of free-form text; your code parses those arguments and executes the function, then feeds the result back to the model so it can continue the conversation with the new information.

This is the foundation of agentic AI: it bridges the gap between a model's reasoning and the outside world. Without function calling, a chatbot can only answer from what it already knows; with function calling, it can query a database for live data, create a record in a CRM, send an email, hit a payment API, or chain multiple actions together to complete a multi-step task.

The standard implementation pattern is a tool loop: send the user message + tool definitions → model returns a tool call → execute the tool → send the result back to the model → repeat until the model returns a final text response. OpenAI, Anthropic, and Google all expose function calling through compatible-ish APIs, and Pydantic-based schema validation has become the de facto pattern for defining tool signatures in Python.