OpenAI Function Calling Generator

Function calling (also called tool use) lets GPT-4o and other models call structured functions instead of generating free text. You define a function schema (name, description, parameters as JSON Schema), and the model decides when to call it based on the conversation. This enables reliable structured output, API integrations, and agentic workflows.

The 'functions' API was the original interface; 'tools' is the modern replacement (introduced mid-2023). Both work the same way, but 'tools' supports multiple tool types (function, code_interpreter, retrieval) and is the recommended interface for all new development. This generator uses the 'tools' format.

In the JSON Schema, list required parameters in the 'required' array. Parameters not listed in 'required' are optional — the model may omit them. For reliable results, make all parameters required and use descriptive enum values or patterns to constrain inputs. Avoid too many optional parameters as the model may hallucinate values.

Supported types: string, number, integer, boolean, array, object, null. Use 'enum' to restrict to specific values. Use 'description' on every parameter — this is critical for model accuracy. Use 'items' for array element type, and 'properties' for nested objects. Avoid deeply nested schemas as they increase token cost and reduce reliability.

After calling chat.completions.create(), check if response.choices[0].message.tool_calls is not None. Extract the function name and parse the JSON arguments string. Execute your function, then append the result as a 'tool' role message. Finally, call the API again with the tool result — the model will incorporate it into its final response.