typing-schema 0.1.1

Convert Python types and typing annotations into a simple schema representation.

Typing schema

Convert Python types and typing annotations (including Annotated, Union, Literal, dataclasses, TypedDicts, and pydantic.BaseModel) into a simple schema representation.

Overview

This library provides utilities to transform Python type hints and function signatures into a JSON-like schema. It is useful for documenting or validating function inputs and the data structures described by Python typing constructs. It can also generate parameter schemas compatible with OpenAI's function-calling API, making it easier to provide accurate function parameter definitions for model function calls.

Key features:

• Convert basic types (str, int, float, bool, list, dict) to simple value schemas.
• Convert Union, Literal, Enum, Annotated (with inline descriptions), dataclasses, TypedDict, and pydantic.BaseModel to object/oneOf/enum schemas.
• Convert a function signature into a schema that lists parameters, types, and required fields.
• Provide extension points: custom type handlers and annotated-doc handlers.

Usage

Basic example using type annotations:

Convert a typing annotation

from typing_schema import typing_to_schema, function_to_schema
from typing import Annotated, Union, Literal

class Record(TypedDict):
    """The record item"""
    id: int
    name: str
    content: Annotated[str | None, "The content (Optional)"]

schema = typing_to_schema(Record)
print(json.dumps(schema, indent=2))

Output:

{
  "type": "object",
  "properties": {
    "id": {
      "type": "integer"
    },
    "name": {
      "type": "string"
    },
    "content": {
      "type": [
        "string",
        "null"
      ],
      "description": "The content (Optional)"
    }
  },
  "required": [
    "id",
    "name"
  ],
  "description": "The record item"
}

Convert a function signature

def func(a: int, b: str = 'x') -> None:
    """Example function"""
    pass

schema = function_to_schema(func)
print(json.dumps(schema, indent=2))

Output:

{
  "type": "object",
  "properties": {
    "a": {
      "type": "integer"
    },
    "b": {
      "type": "string",
      "default": "x"
    }
  },
  "required": [
    "a"
  ],
  "description": "Example function"
}

Extension points

• type_handler: pass a callback to typing_to_schema or function_to_schema to handle custom types. If it returns a schema the converter will use it.
• annotated_doc_handler: pass a callback to extract documentation from Annotated[...] metadata.

Handle custom types

def handler(type) -> dict | None:
    if (
        type is datetime.datetime
        or isinstance(type, datetime.date)
        or issubclass(type, datetime.date)
    ):
        return {"type": "string", "description": "Handled string type"}
    return None

schema = typing_to_schema(datetime.date, type_handler=handler)
print(json.dumps(schema, indent=2))

Output:

{
  "type": "string",
  "description": "Handled string type"
}

Extract documentation from Annotated metadata

By default, the first str value in Annotated metadata is used as the description.

class Doc:
    def __init__(self, value: str):
        self.value = value

def doc_handler(args: tuple[type, ...]) -> str | None:
    for arg in args:
        if isinstance(arg, Doc):
            return arg.value
    return None

class Record(TypedDict):
    name: Annotated[str, Doc("The name of the person")]

schema = typing_to_schema(Record, annotated_doc_handler=doc_handler)
print(json.dumps(schema, indent=2))

output:

{
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "The name of the person"
    }
  },
  "required": [
    "name"
  ]
}