OA Python SDK

oapython is the native Python distribution for the Rust OA runtime. It installs the oa import package and mirrors Rust's public module structure:

python -m pip install oapython

import oa

engine = oa.Engine()
one = oa.matrix.ones(engine, [2, 3])
two = oa.matrix.full(engine, [2, 3], 2.0)
total = oa.matrix.add(one, two)

assert total.shape == [2, 3]
assert total.to_list() == [3.0] * 6

image = oa.Image.from_matrix(oa.matrix.ones(engine, [3, 2, 2]), "chw", "rgb")
flipped = oa.image.flip(image, horizontal=True, vertical=False)

Python mirrors Rust module ownership: root value aliases remain concise, stateless operations live in oa.matrix, and the runtime owner is explicit. Callers may use ordinary local abbreviations such as import oa.matrix as oam or import oa.core as oac; OA does not publish duplicate alias modules. Host observation through read_f32() or to_list() is the synchronization boundary.

The canonical PyO3 implementation lives in ../../src/py. Root ../../pyproject.toml and ../../uv.lock own packaging and environment metadata; this SDK directory owns the importable package, PyO3 crate, and tutorials.

Build from the repository root with:

python3 tool/build/workflow.py --profile debug --python-only

The workflow creates or reuses the root .venv, rebuilds the native extension, then verifies package imports and native/distribution version agreement in a fresh Python process. Re-run it after changing native exports; an older installed extension does not gain renamed classes from edits to the Python wrappers.

Tests are located in the repository test/py/ directory and run with:

.venv/bin/python -m unittest discover -s test/py -v

Build the portable Python 3.10+ Linux wheel with:

.venv/bin/pip install ziglang
PATH="$PWD/.venv/bin:$PATH" ZIG_COMMAND=python-zig .venv/bin/maturin build --release --zig

The explicit Zig build prevents a developer workstation's newer glibc symbols from leaking into a wheel labeled for PyPI.

The Rust-first preview exposes runtime ownership and logging, explicit events, typed dense matrices, Matrix and ML functional operations, Image values/codecs/transforms, Audio values/codecs/DSP/capture/playback, cryptography, vision metrics, and the current Video surface.

Neural-network modules, optimizers, training sessions and programs, callbacks, reinforcement-learning trainers, flow matching, and SDK NLP recipes remain experimental. Python targets compute, ML, media, Plot, and Viewer preview; low-level Vulkan and UI construction remain Rust-only. Capability admission and hardware qualification are separate from binding availability.