function calling

Function calling (Tool Calling) empowers Large Language Models (LLMs) to generate structured JSON, enabling reliable interaction with external APIs and real-time data access.

Function calling provides a robust mechanism for connecting LLMs (e.g., GPT-4, Gemini) to external systems, significantly extending their capabilities beyond their training data. The workflow is precise: a developer defines available tools using a JSON schema (e.g., `get_stock_price(ticker)`). When a user asks a query like, 'What is the price of TSLA?', the model intelligently responds with a structured object: `{"name": "get_stock_price", "arguments": {"ticker": "TSLA"}}`. The application code executes this function call, retrieves the real-world result, and sends that output back to the model. The LLM then synthesizes the final, accurate, and context-aware answer for the user. This multi-step process is essential for building data-aware and action-oriented AI agents.

https://platform.openai.com/docs/guides/function-calling