Blazing-fast Python ASGI¶
70–95% lower latency at p50–p99
Switch the server, not your app. h2corn runs FastAPI, Starlette, Django, Litestar and any ASGI 3 app, with end-to-end HTTP/2 and a production worker supervisor.
Fast by default
HTTP framing, TLS, and stream multiplexing run natively. Requests only cross into Python when there is real handler work to do.
Secure deployments
HTTP/2 end-to-end keeps the proxy → app hop off HTTP/1.1, removing the downgrade surface that request-smuggling research targets.
Operator-friendly
A multi-worker supervisor built for long-running deployments — rolling reload, live scaling, recycling, heartbeats.
Why h2corn¶
A browser reaches your edge over HTTP/3 or HTTP/2. What happens on the next hop is almost always HTTP/1.1 — the proxy translates the request down before the application server ever sees it. That translation is what HTTP/2 downgrading research targets, and it is the hop most deployments never think about.
h2corn removes it. The proxy speaks h2c straight into the application, with WebSockets on that same connection (RFC 8441) instead of a hijacked HTTP/1.1 socket. The edge keeps doing what edges are good at; the last hop stops being the weak one.
Speed comes with it. In most Python services the handler is quick and the server around it is what keeps p99 high, so moving connections, TLS and HTTP into Rust leaves the event loop doing one thing — running your code. Across the twenty comparable scenarios of the benchmark suite, the median one comes out at 84% lower p50 and 82% lower p99 than the fastest of uvicorn, hypercorn and gunicorn in that same scenario; four workers serve a small plaintext GET at ~242k RPS, p99 0.8 ms, about 5× the nearest alternative on the identical Starlette app and over 20× on HTTP/2. See benchmarks.