PyBridge

End-to-end type safety from Python to TypeScript — no codegen ceremony, no schema drift.

Define procedures in Python with standard type hints + Pydantic. Run one command. Get a fully typed TypeScript client.

pip install pybridge-rpc
pybridge generate --bridge examples.basic:bridge --out client/api.ts --hooks

# server.py
from pybridge import Bridge
from pydantic import BaseModel

bridge = Bridge()

class User(BaseModel):
    id: str
    name: str
    email: str

@bridge.procedure("users.create")
async def create_user(input: User) -> User: ...

app = bridge.asgi()

import { createClient, type AppRouter } from "./api";
const api = createClient<AppRouter>("http://localhost:8000");
const user = await api.users.create({ id: "1", name: "Kofi", email: "k@example.com" });
//    ^ fully typed as User

Features

• Zero-ceremony typed RPC — Python function signature is the API contract. No .input()/.output() wrappers, no separate schema files.
• Pydantic → TypeScript — full type projection: nested models, unions, literals, enums, datetime, UUID, tuples, dict, optional/nullable.
• One-command codegen — pybridge generate emits a single api.ts with types + proxy client + optional TanStack Query hooks.
• ASGI transport — HTTP, WebSocket subscriptions, and SSE streams out of the box (built on Starlette).
• Framework integrations — drop into FastAPI, Django, Sanic, or Litestar without rewriting your app.
• File uploads, CORS & CSRF helpers, observer protocol for logging/metrics/tracing hooks.
• OpenAPI 3.x exporter — get a spec for tooling that needs one, without making it the source of truth.
• Watch mode — watchfiles-backed regeneration with mtime-polling fallback.
• Typed errors — ProcedureError codes propagate to the client.
• Fully typed package — ships py.typed.

Why PyBridge

Full-stack teams with a Python backend and a TypeScript frontend live with a constant gap: define models in Pydantic, redefine them as TS interfaces, hope they stay in sync. The usual fixes have rough edges:

The goal is the same DX tRPC/oRPC developers enjoy — with Python as the source of truth. See docs/design.md for the full rationale.

Status

Phases 1–3 are complete and tested (33 tests passing). Currently 0.2.x — beta. API is stable for the documented surface; breaking changes will be called out in release notes.

Documentation

All docs live in docs/:

• Design — the problem, approach, architecture.
• Quickstart — install, server + client, feature list.
• Streaming — WebSocket subscriptions and SSE streams.
• Authentication — bearer tokens, cookie sessions, CSRF.
• Observability — observers, timeouts, body limits.
• Framework integrations — FastAPI, Django, Sanic, Litestar.
• TanStack example — Router + Query end-to-end.
• Performance — benchmarks.
• Project layout — module map.
• Roadmap — phases, tech stack, status.

License

MIT — see LICENSE.