toncorn

An ASGI web server for free-threaded Python, powered by tonio.

toncorn is a fork of uvicorn that swaps the asyncio runtime for tonio.

Install

$ pip install toncorn

toncorn requires free-threaded CPython 3.14 or newer (python3.14t) and runs PYTHON_GIL=0. Linux and macOS only; Windows is not supported and won't be (tonio doesn't target it).

For the optional protocol/parser extras (matching upstream uvicorn's [standard] set):

$ pip install 'toncorn[standard]'

Quickstart

example.py:

async def app(scope, receive, send):
    assert scope["type"] == "http"
    await send({
        "type": "http.response.start",
        "status": 200,
        "headers": [(b"content-type", b"text/plain")],
    })
    await send({
        "type": "http.response.body",
        "body": b"Hello, world!",
    })

Run it:

$ PYTHON_GIL=0 toncorn example:app

You'll see a banner identifying the toncorn version and the uvicorn version it tracks:

INFO:     toncorn 0.48.0.0 (uvicorn 0.48.0)
INFO:     Started server process [12345]
INFO:     Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)

The package ships both import toncorn and import uvicorn — the source tree lives under the original uvicorn package name to keep the diff against upstream small, and toncorn is a thin re-export. Pick whichever import you prefer; they expose the same Config, Server, run, main.

Note: this means toncorn and the real PyPI uvicorn collide on the uvicorn import name; don't install both in the same environment.

Programmatic use

import toncorn
toncorn.run("example:app", host="127.0.0.1", port=8000)

Or with an explicit Config + Server:

import toncorn

config = toncorn.Config(app=app, host="127.0.0.1", port=8000)
server = toncorn.Server(config)
server.run()

Differences from uvicorn

• Runtime is tonio, not asyncio. There's no --loop flag, no uvloop/asyncio switch. Everything runs on a single tonio runtime with a configurable thread pool.
• Concurrency model is threads, not processes. The old --workers N flag is replaced by --threads N, which sizes the tonio scheduler's thread pool. No multiprocessing supervisor, no preforked workers, no graceful-restart process orchestration.
• No --reload. The auto-reload supervisor is gone. Use your editor's hot-reload or a separate process manager during development.
• No Windows support. POSIX only.
• No in-tree WSGI middleware. The deprecated WSGI bridge has been removed; use a2wsgi or asgiref.wsgi if you need WSGI compatibility.
• Python 3.14+ only. Earlier Python versions are not supported.

The version number reflects both upstream and fork: {uvicorn}.{toncorn}, so 0.48.0.0 tracks uvicorn 0.48.0 with toncorn patch 0. Patch-level releases against the same uvicorn baseline bump the last segment (0.48.0.0 → 0.48.0.1).

Upstream tracking

The source tree under uvicorn/ is kept as close to upstream uvicorn as possible to make merges/rebases tractable. Fork-specific changes are intentionally minimal and called out where they appear. The toncorn/ package is a re-export shim plus the toncorn-specific version.

License

BSD-3-Clause, same as upstream uvicorn. See LICENSE.md.