PyRe 1.0 · Available Now · Windows x64

Your Python.
A steadier run.

Keep your workload on your machine. PyRe brings resource planning, conservative routing, a durable controller, and run evidence to local Python execution. Run verified CPU and NVIDIA CUDA jobs with output parity and receipt logging.

The current desktop release runs CPU and NVIDIA CUDA jobs. Intel XPU and AMD ROCm can be probed, but cannot currently be scheduled. Their setup links and example branches below are reference material for future support, not runnable desktop options in this release.

Plan the run. Follow the evidence.

Bring your own workload.

PyRe is an execution controller, not a chat model or a security sandbox. Run only code and commands you trust. It does not rewrite ordinary Python or move data to a GPU for you.

The PySide6 desktop submits work to the same durable local controller used for resource admission, execution, status, recovery, and receipts. Current protected host acceptance passed on Windows x64 with CPython 3.10, 3.11, 3.12, and 3.13. Actual CPU/CUDA execution and output parity passed on an NVIDIA GeForce RTX 5070 Ti. These results describe the tested configurations, not every GPU or workload.

Optional acceleration still depends on compatible code, the operating system, an officially supported device and driver, and a matching framework build. PyRe 1.0 is available now for Windows x64 with perpetual one-time purchase at $29 USD.

Run code on the device you selected.

PyRe reserves and launches the selected lane. Your workload must explicitly create or move tensors on that device. Start with the downloadable CPU/GPU proof script; set its BACKEND constant to match the mode you choose in PyRe. Its CPU path uses only the Python standard library, while its optional accelerator path imports PyTorch, checks the real runtime device, includes transfers and synchronization in timing, compares output with CPU, and exits with an error instead of silently falling back.

1. Create a separate virtual environment for the accelerator and install both PyRe and the framework build that matches the operating system, driver, and device. Use the current official PyTorch installer selector. Intel users should also follow the official PyTorch XPU guide. Windows AMD users must check the exact GPU, Python, PyTorch, and ROCm combination in AMD's official Windows support matrix.
2. Before starting the controller, set the matching absolute interpreter path: PYTHON_GPU_CUDA_PYTHON, PYTHON_GPU_XPU_PYTHON, or PYTHON_GPU_ROCM_PYTHON. The current desktop Settings dialog shows these effective paths; it does not install frameworks or change them for a controller that is already running.
3. Set BACKEND near the top of how-to.py to "cpu", "cuda", "xpu", or "rocm". In PyRe, add that file, choose the matching mode and physical device, and select Check backend. A backend check proves current framework/device identity, not that arbitrary code will accelerate.
4. Select Run selected. In the controller output and receipt, confirm the requested and selected backend. The example also prints actual_device, parity=PASS, and transfer-inclusive elapsed time. AMD ROCm PyTorch intentionally uses the torch.cuda Python API and verifies that the runtime reports HIP.

# PowerShell — replace this placeholder with the absolute GPU-environment interpreter
$env:PYTHON_GPU_CUDA_PYTHON = "C:\path\to\gpu-venv\Scripts\python.exe"
& ".\Launch PyRe.cmd"

Start PyRe from that same PowerShell session so the controller inherits the CUDA setting. The XPU and ROCm variables are reference configuration only; they do not enable scheduling in the current desktop release.

Choose an explicit accelerator mode when GPU execution is required. Auto keeps ambiguous work on CPU. PyRe does not promise OpenCL, DirectML, Vulkan, Apple GPU support, or automatic conversion of CPU code.

Accelerator status

Release 1.0.0 is available for Windows x64.

PyRe 1.0 is published and available for immediate purchase and download. The native Windows x64 setup package is built with AES-256 encrypted release distribution and authenticated via Hugging Face and Fly.io.

A one-time purchase of $29 USD via Stripe or PayPal entitles you to perpetual use of PyRe with no monthly fees. Immediately after purchase, a secure download link and purchase receipt are delivered to your email.