Beyond the Transistor

Install the toolkit

Install the Python package with the CPU emulator so you can develop without hardware, or install the default build on a machine with a Q.ANT card.

The CPU emulator build from source was tested with toolkit 2.3.1 and Python 3.13 on Linux on 8 October 2026. The toolkit's own tests passed (109 passed, 1 deselected).

The toolkit has two builds. Choose the one that matches your machine:

  • CPU emulator (cpu-backend). Runs on any Linux machine. You don’t need a Q.ANT card or driver. Use it to learn the API and run the examples.
  • Default. Links against the Q.ANT driver and sends work to a physical card.

The toolkit is not on PyPI yet. A user asked for it in issue #4, and a Q.ANT collaborator replied that making the toolkit available on a package manager is on the roadmap.

Before you begin

  • Install Python 3 and pip. We used Python 3.13.
  • Install Git.
  • To build from source, install a Rust toolchain with rustup. The repository’s rust-toolchain.toml pins Rust 1.93.0, and rustup installs that version on the first cargo run.

Install the CPU emulator from source

This path builds a wheel with the emulator compiled in.

  1. Clone the repository with its submodules:

    git clone --recurse-submodules https://github.com/Q-ANT-GmbH/qant_native_computing_toolkit.git

  2. Go to the repository folder:

    cd qant_native_computing_toolkit

  3. Install maturin, the build tool for Rust Python extensions:

    pip install maturin

  4. Build a release wheel with the CPU backend:

    maturin build --release -F cpu-backend -o wheels

  5. Install the wheel:

    pip install wheels/*.whl

Install for development

If you plan to change the toolkit code, the README recommends an editable development install instead of a wheel:

pip install maturin
maturin develop -F cpu-backend

maturin develop builds the extension and installs it into the active Python environment.

Install with Q.ANT hardware

  1. Install the Q.ANT drivers for your card. The toolkit README lists this as a prerequisite for the default build.

  2. Install the Python package. Use one of these options:

    • From a release wheel:

      pip install qant_native_computing_toolkit.<version>.whl

    • From source, in the cloned repository:

      pip install .

  3. For C or C++, install the Debian package:

    apt install ./qant_native_computing_toolkit.<version>.deb

To build the C and C++ shared library from source, run cargo build --release. Add -F cpu-backend for a standalone build without the driver:

cargo build --release -F cpu-backend

We did not test the hardware path. We have no access to a Q.ANT card.

Release packages

The releases page listed v2.3.1 as the latest release when we checked on 8 October 2026. Its assets were six zip archives plus the source code:

AssetContents
qant-native-computing-toolkit-wheels-default-v2.3.1.zipPython wheels, default build
qant-native-computing-toolkit-wheels-cpu-backend-v2.3.1.zipPython wheels, CPU emulator build
qant-native-computing-toolkit-deb-default-v2.3.1.zipDebian package, default build
qant-native-computing-toolkit-deb-cpu-backend-v2.3.1.zipDebian package, CPU emulator build
qant-native-computing-toolkit-python-api-doc-v2.3.1.zipPython API documentation
qant-native-computing-toolkit-c-api-doc-v2.3.1.zipC API documentation

The wheels are inside the zip archives, so the releases page doesn’t show which platforms or Python versions they cover. Download the archive and check the wheel file names before you install. The README says to install “the package wheel for your python version”.

Verify the installation

  1. Check that Python finds the toolkit and an NPU:

    python -c "import qant_native_computing_toolkit as q; print(q.generic.get_available_npus())"

    On the CPU emulator this prints:

    {0: 'cpu_backend'}

    On hardware the dictionary maps each device ID to the card’s serial number.

  2. Check which backend you are using:

    python -c "import qant_native_computing_toolkit as q; print(q.info.get_driver_info())"

    On the CPU emulator this prints:

    cpu-backend;commit:none;

Run the toolkit’s tests

The toolkit ships its own pytest suite in python/tests.

  1. Go to the tests folder:

    cd python/tests

  2. Run the tests, excluding the logging test:

    python -m pytest -q -m "not logging"

On the CPU emulator the result was 109 passed, 1 deselected. The tests need PyTorch, which the toolkit’s CONTRIBUTING guide installs from the CPU-only index.

What’s next