Types

@wishful.type registers schemas that get inlined into the prompt. Works with dataclasses, Pydantic models, and TypedDicts.

Think of it as teaching the LLM your native language.

import wishful
from dataclasses import dataclass

@wishful.type
@dataclass
class Book:
    title: str
    author: str
    year: int

Bind a type to specific function outputs:

from typing import TypedDict

class ProductInfo(TypedDict):
    name: str
    price: float

wishful.type(ProductInfo, output_for=["fetch_product", "create_product"])

Pydantic with constraints (v2-compatible):

from pydantic import BaseModel, Field

@wishful.type(output_for="plan_trip")
class TripPlan(BaseModel):
    city: str
    days: int = Field(ge=1, le=30)
    vibe: str = Field(description="tone of the itinerary")

What happens:

• Definitions get serialized to Python code and injected into the system prompt.
• Output bindings are passed to the LLM so it knows which functions must return which type.
• Docstrings travel too, so you can nudge style (“sound like a pirate”, “Yoda-speak”, etc.). Yes, really.

See examples/07_typed_outputs.py and the 30 tests in tests/test_types.py for more patterns.