pythscribe — @wasm for ordinary Python

One decorator. Your function stays Python. It runs compiled — in the browser tab, in-process on the server (sandboxed, GIL-free, bit-for-bit), or as plain Python when nothing is built — and you can always tell which.

pip install pythscribe            # the decorator, the loader, the fallback, both framework adapters
pip install pythscribe[server]    # + wasmtime: run kernels in-process on the server
pip install pythscribe[gradio]    # + Gradio (the custom component ships inside the wheel)

from pythscribe import wasm

@wasm
def dot(xs: list[float], ys: list[float]) -> float:
    acc = 0.0
    for i in range(len(xs)):
        acc = acc + xs[i] * ys[i]
    return acc

python -m pythscribe.build kernels.py   # compile each @wasm def with `pyths` -> __pythscribe__/<fn>/

Without the build step the function is still your function (plain Python). With it, the same call runs the compiled kernel, and the decorator tells you which path ran:

>>> from pythscribe import binding_of
>>> binding_of(dot).mode          # 'server' | 'browser' | 'fallback'
'server'

Documentation & demos

• Installation, supported platforms, and the full guide — see the project README (Installation section).
• Runnable demos (Gradio and Streamlit, browser + server) — demos/.
• Worked examples (@wasm kernels, image preprocessing, typed arrays) — examples/.

Licence

The pip package is MIT (pythscribe/LICENSE). The wheel bundles the pyths compiler binary (pythscribe/_bin/, FSL-1.1-ALv2), so the wheel's licence expression is MIT AND LicenseRef-FSL-1.1-ALv2; its .dist-info/licenses/ carries both texts, the vendored runtime's notice, a per-installed-path map, and the third-party notices of the crates linked into the binary. Compiled output of your own code is yours.