dyadpy 0.1.1

A type-safe RPC bridge between Python and TypeScript. The function signature is the contract.

dyadpy (Python)

A type-safe RPC bridge between Python and TypeScript.

uv add 'dyadpy==0.1.0a0'   # alpha — drop the pin once v0.1.0 ships

This is the Python half of Dyadpy. It ships:

• A thin ASGI framework (dyadpy.App) that uses your function signatures as the contract — no separate Pydantic models declared above the handler.
• A type extractor that walks inspect.signature + typing.get_type_hints, normalizes through msgspec's native JSON Schema export, and produces a canonical IR.
• A codegen that turns the IR into a single client.ts for your frontend.
• A CLI (dyadpy dev, dyadpy build, dyadpy codegen, dyadpy init) that runs the whole loop in one process.

For the full story, the design rationale, and a side-by-side comparison vs. FastAPI + openapi-typescript / tRPC / Encore.ts / Connect-RPC, see the repo README.

30-second example

from dyadpy import App, stream, raises
from dataclasses import dataclass
import msgspec

app = App()

class CreatePost(msgspec.Struct):
    title: str
    body: str

class Post(msgspec.Struct):
    id: int
    title: str
    body: str

@dataclass
class PostNotFound(Exception):
    post_id: int

@app.post("/posts")
async def create_post(data: CreatePost) -> Post: ...

@app.get("/posts/{post_id}")
@raises(PostNotFound)
async def get_post(post_id: int) -> Post: ...

class Tick(msgspec.Struct, tag="tick"):
    seq: int

@app.get("/ticks")
async def ticks(count: int) -> stream[Tick]:
    for i in range(count):
        yield Tick(seq=i)

Run it:

dyadpy dev server.app:app

The watcher writes frontend/src/lib/dyadpy/client.ts automatically. Then in your frontend:

import { api } from "@/lib/dyadpy/client";

const post = await api.createPost({ data: { title: "hi", body: "world" } });

for await (const ev of api.ticks({ count: 10 })) {
  /* typed */
}

Primitives in this package

Primitive	Purpose
App + @app.{get,post,put,patch,delete}	Route decorators.
Annotated[T, Body / Query / Path / Header / Cookie / File / Form]	Parameter location markers.
Annotated[list[T], Query()]	Repeated query params (?tag=a&tag=b).
Bytes	Raw request / response bodies. Skips the JSON envelope.
stream[T]	Typed SSE — client gets AsyncIterable<T>.
@raises(E1, E2, …)	Typed error union → Result<T, E1 | E2> on the TS side.
Context.set_status / set_header / set_cookie / after	Shape the response without dropping to Starlette.
Depends(provider)	DI in the FastAPI shape.
after(fn, …)	Run a callback after the response is sent.
InMemoryBackend + TaskBackend Protocol	Background jobs.
dyadpy.otel.instrument(app)	One OpenTelemetry span per request (dyadpy[otel]).
dyadpy openapi / swift / kotlin (CLI)	Emit OpenAPI 3.1, Swift, or Kotlin clients off the same IR.

Full reference: https://github.com/tamimbinhakim/dyadpy/blob/main/docs/reference.md

Optional extras

uv add 'dyadpy[pydantic]'  # Pydantic plugin (model_validate + model_json_schema)
uv add 'dyadpy[otel]'      # OpenTelemetry middleware
uv add 'dyadpy[all]'       # everything

Scope

Dyadpy ships at the wire level: RPC, typed streaming, typed errors, cancellation, file uploads, dependency injection. It does not ship vertical integrations — no LLM types, no React hooks in core, no chat-bot primitives. Those layers compose on top of the fundamentals and live in their own packages.

License

MIT