Could not make a good title … better see the code.

I’m trying to use pydantic to process nested configuration data. Of course, I want to validate the whole structure, but the final result should be a class at the top level (BaseModel or dataclass), but the fields should be dicts.

from pydantic import BaseModel, Field

class Inner(BaseModel):
    name: str 
    count: int = Field(default=123)

class Outer(BaseModel):
    main: Inner
    aux: str = Field(default="stable")

DATA = {'main': {'name': 'blues'}}
mclass = Outer.validate(DATA)

# class -> attr based access
print(mclass.main.count)

# dict -> key based access
mdict = mclass.model_dump()
print(mdict['main']['count'])

# what I want: class at the top level, but field values as dict
mxxx.main['count'] # <--- ???

I could change the type of all fields corresponding to nested models:

mclass.main = mclass.main.model_dump()

But can the pydantic define such nested model “naturally”? I’m not an experienced user. I tried the TypedDict (as in this related post: /a/74643914/5378816), but it cannot create defaults or use custom validators.