OpenAI Functions Calling

OpenAI shipped the first mainstream function calling API, and its conventions shaped how the whole industry does tools. In the current API you pass a tools array where each entry has a name, description, and JSON Schema parameters. When the model decides to act, the response contains one or more tool calls — each with an ID, a function name, and an arguments field that arrives as a JSON string you must parse. You execute the function, return the output tied to the matching call ID, and the model continues.

This API is worth knowing in detail even if you build on another provider, because it’s the de facto standard: many open-source inference servers, gateways, and frameworks expose OpenAI-compatible endpoints, so this exact shape is what you’ll meet everywhere. Distinctive features include strict: true mode, which constrains generation so arguments are guaranteed to match your schema, and tool_choice for forcing a specific function, requiring some tool, or disabling tools entirely.

In practice you’ll define tools in the SDK, call the model, then loop: parse arguments with json.loads, dispatch to your function, and append a result message referencing the tool_call_id before calling again. You’ll handle parallel tool calls (several in one response) and decide between the older Chat Completions endpoint and the newer Responses API, which wraps the same tool semantics with server-side state and built-in tools like web search and code execution.