Hatchway is an API framework inspired by the likes of FastAPI, but while trying to keep API views as much like standard Django views as possible.

It was built for, and extracted from, Takahē; if you want to see an example of it being used, browse its api app.

Install Hatchway from PyPI:

pip install django-hatchway

And add it to your INSTALLED_APPS:

INSTALLED_APPS = [
    ...
    "hatchway",
]

To make a view an API endpoint, you should write a standard function-based view, and decorate it with @api_view.get, @api_view.post or similar:

from hatchway import api_view

@api_view.get
def my_api_endpoint(request, id: int, limit: int = 100) -> list[str]:
    ...

The types of your function arguments matter; Hatchway will use them to work out where to get their values from and how to parse them. All the standard Python types are supported, plus Pydantic-style models (which ideally you should build based on the hatchway.Schema base class, as it understands how to load things from Django model instances).

Your return type also matters - this is what Hatchway uses to work out how to format/validate the return value. You can leave it off, or set it to Any, if you don't want any return validation.

You add API views in your urls.py file like any other view:

urlpatterns = [
    ...
    path("api/test/", my_api_endpoint),
]

The view will only accept the method it was decorated with (e.g. GET for api_view.get).

If you want to have two or more views on the same URL but responding to different methods, use Hatchway's methods object:

from hatchway import methods

urlpatterns = [
    ...
    path(
        "api/post/<id>/",
        methods(
            get=posts.post_get,
            delete=posts.posts_delete,
        ),
    ),
]

There are four places that input arguments can be sourced from:

• Path: The URL of the view, as provided via kwargs from the URL resolver
• Query: Query parameters (request.GET)
• Body: The body of a request, in either JSON, formdata, or multipart format
• File: Uploaded files, as part of a multipart body

By default, Hatchway will pull arguments from these sources:

• Standard Python singular types (int, str, float, etc.): Path first, and then Query
• Python collection types (list[int], etc.): Query only, with implicit list conversion of either one or multiple values
• hatchway.Schema/Pydantic BaseModel subclasses: Body only (see Model Sourcing below)
• django.core.files.File: File only

You can override where Hatchway pulls an argument from by using one of the Path, Query, Body, File, QueryOrBody, PathOrQuery, or BodyDirect annotations:

from hatchway import api_view, Path, QueryOrBody

@api_view.post
def my_api_endpoint(request, id: Path[int], limit: QueryOrBody[int] = 100) -> dict:
    ...

While Path, Query, Body and File force the argument to be picked from only that source, there are some more complex ones in there:

• PathOrQuery first tries the Path, then tries the Query (the default for simple types)
• QueryOrBody first tries the Query, then tries the Body
• BodyDirect forces top-level population of a model - see Model Sourcing, below.

When you define a hatchway.Schema subclass (or any other pydantic model subclass), Hatchway will presume that it should pull it from the POST/PUT/etc. body.

How it pulls it depends on how many body-sourced arguments you have:

• If you just have one, it will feed it the top-level keys in the body data as its internal values.
• If you have more than one, it will look for its data in a sub-key named the same as the argument name.

For example, this function has two body-sourced things (one implicit, one explicit):

@api_view.post
def my_api_endpoint(request, thing: schemas.MyInputSchema, limit: Body[int] = 100):
    ...

This means Hatchway will feed the schemas.MyInputSchema model whatever it finds under the thing key in the request body as its input, and limit will come from the limit key.

If limit wasn't specified, then there would be only one body-sourced item, and Hatchway would feed schemas.MyInputSchema the entire request body as its input.

You can force a schema subclass to be fed the entire request body by using the BodyDirect[MySchemaClass] annotation on its type.

The return value of an API view, if provided, is used to validate and coerce the type of the response:

@api_view.delete
def my_api_endpoint(request) -> int:
    ...

It can be either a normal Python type, or a hatchway.Schema subclass. If it is a Schema subclass, the response will be fed to it for coercion, and ORM objects are supported - returning a model instance, a dict with the model instance values, or an instance of the schema are all equivalent.

A typechecker will honour these too, so we generally recommend returning instances of your Schema so that your entire view benefits from typechecking, rather than relying on the coercion. You'll get typechecking in your Schema subclass constructors, and then typechecking that you're always returnining the right things from the view.

You can also use generics like list[MySchemaClass] or dict[str, MySchemaClass] as a response type; generally, anything Pydantic allows, we do as well.

If you want to do more to your response than just sling some data back at your client, you can return an ApiResponse object instead of a plain value:

from hatchway import api_view, ApiResponse

@api_view.delete
def my_api_endpoint(request) -> ApiResponse[int]:
    ...
    return ApiResponse(42, headers={"X-Safe-Delete": "no"})

ApiResponse is a standard Django HTTPResponse subclass, so accepts almost all of the same arguments, and has most of the same methods. Just don't edit its .content value; if you want to mutate the data you passed into it, that is stored in .data.

Note that we also changed the return type of the view so that it would pass typechecking; ApiResponse accepts any response type as its argument and passes it through to the same validation layer.

Hatchway allows you to say that Schema subclasses can pull their values from individual query parameters or body values; these are normally flat strings, though, unless you're looking at a JSON-encoded body, or multiple repeated query parameters.

However, it will respect the use of name[] to make lists, and name[key] to make dicts. Some examples:

• A a=Query[list[int]] argument will see url?a=1 as [1], url?a=1&a=2 as [1, 2], and url?a[]=1&a[]=2 as [1, 2].
• A b=Body[dict[str, int]] argument will correctly accept the POST data b[age]=30&b[height]=180 and give you {"age": 30, "height": 180}.

These will also work in JSON bodies too, though of course you don't need them there; nevertheless, they still work for compatibility reasons.