JSON Schema → OpenAI Function

OpenAI tool definitions live in the tools array of your chat completions request. Each entry is a wrapper around a function spec, which itself contains a JSON Schema for the parameters. Adding strict: true activates structured outputs, which forces the model to emit arguments that parse and conform to your schema.

What does <code>strict: true</code> do?

It tells OpenAI to enforce the schema on the model's output, guaranteeing that tool_call.function.arguments parses as valid JSON conforming to your schema. Strict mode requires every property to be in required and additionalProperties: false at every object level.

Why does strict mode reject some schemas?

OpenAI's structured-outputs implementation supports a subset of JSON Schema. It rejects oneOf/not, $ref, recursive schemas, integers larger than i64, and a few other edge cases. The error message points to the offending node.

Is the wrapping <code>{"type": "function", "function": ...}</code> still required?

For the Chat Completions API, yes. The Responses API uses a flatter shape. The OpenAI SDK abstracts this — most folks paste the wrapped form into tools.

Where do parameter descriptions go?

Inside each property ("description": "..."). The model uses these heavily when deciding which tool to call and how to fill arguments. Don't skip them.

• Forgetting additionalProperties: false at every object level when using strict mode.
• Putting required: ["foo", "bar"] at the top level but only listing one of them in properties.
• Using $ref or recursive types — strict mode rejects them.
• Skipping per-property description; the model picks tools based on these and quality drops without them.

• JSON Schema → Anthropic Tool

  Wrap a JSON Schema as an Anthropic tool definition for the messages API.

• OpenAPI → Tool Definitions

  Extract tool definitions from an OpenAPI 3.x spec for OpenAI or Anthropic.