diff --git a/src/hip/__init__.pyi b/src/hip/__init__.pyi
index e69de29b..2f7232a9 100644
--- a/src/hip/__init__.pyi
+++ b/src/hip/__init__.pyi
@@ -0,0 +1 @@
+from .models import Request, SyncRequestData, AsyncRequestData
diff --git a/src/hip/models.pyi b/src/hip/models.pyi
new file mode 100644
index 00000000..d0176b69
--- /dev/null
+++ b/src/hip/models.pyi
@@ -0,0 +1,81 @@
+import typing
+
+HeadersType = typing.Union[
+    typing.Mapping[str, str],
+    typing.Mapping[bytes, bytes],
+    typing.Iterable[typing.Tuple[str, str]],
+    typing.Iterable[typing.Tuple[bytes, bytes]],
+]
+
+class SyncRequestData:
+    """Represents a synchronous request data object. Basically a wrapper around whatever
+    a user passes in via 'data', 'files', 'json', etc parameters. We can take bytes /
+    strings, iterators of bytes or strings, and files. We can also sub-class / implement
+    a file-like interface for things like multipart-formdata so it's possible to
+    rewind (unlike the current urllib3 interface).
+
+    Maybe in the future we can expose details about when an object can be sent via
+    'socket.sendfile()', etc. This would have to be communicated somehow to the
+    low-level backend streams.
+
+    When we're handed a file-like object we should take down the starting point
+    in the file via .tell() so we can rewind and put the pointer back after
+    the seek() call in '.content_length'
+
+    Ref: https://github.com/python-trio/urllib3/issues/135
+    """
+
+    def read(self, nbytes: int) -> bytes:
+        """Don't know what to call this? Grab some data from the pool of data
+        to be sent across the wire.
+        """
+    @property
+    def content_length(self) -> typing.Optional[int]:
+        """We can get a proper content-length for bytes, strings, file-like objects
+        that are opened in binary mode (is that detectable somehow?). If we hand
+        back 'None' from this property it means that the request should use
+        'Transfer-Encoding: chunked'
+        """
+    def rewind(self) -> None:
+        """This function rewinds the request data so that it can be retransmitted
+        in the case of a timeout/socket error on an idempotent request, a redirect,
+        etc so that the new request can be sent. This works for file-like objects
+        and bytes / strings (where it's a no-op).
+        """
+    @property
+    def is_rewindable(self) -> bool:
+        """We should return a bool whether .rewind() will explode or not."""
+
+class AsyncRequestData(SyncRequestData):
+    """The same as above except also accepts async files (we'll have to
+    handle all the possible APIs here?) and async iterators.
+    """
+
+    async def read(self, nbytes: int) -> bytes: ...
+    async def rewind(self) -> None: ...
+
+class Request:
+    """Requests aren't painted async or sync, only their data is.
+    By the time the request has been sent on the network and we'll
+    get a response back the request will be attached to the response
+    via 'SyncResponse.request'. At that point we can remove the 'data'
+    parameter from the Request and only have the metadata left so
+    users can't muck around with a spent Request body.
+
+    The 'url' type now is just a string but will be a full-featured
+    type in the future. Requests has 'Request.url' as a string but
+    we'll want to expose the whole URL object to do things like
+    'request.url.origin' downstream.
+
+    Also no reason to store HTTP version here as the final version
+    of the request will be determined after the connection has
+    been established.
+    """
+
+    def __init__(
+        self,
+        method: str,
+        url: str,
+        headers: HeadersType,
+        data: typing.Union[SyncRequestData, AsyncRequestData]=None,
+    ): ...