Stethos 0.1.0 — desktop installers

Stethos is an offline desktop app for respiratory diagnostics. It takes breath sounds, a chest X-ray, or spirometry data, classifies each one, and fuses the findings into a risk assessment with a per-modality breakdown. Reports export as PDF, FHIR, DICOM-SR, or JSON. All inference runs locally — nothing leaves the machine.

This repository mirrors the installers published on the GitHub release. It exists because the CUDA packages are ~3 GB each, which is larger than GitHub's 2 GB per-asset limit.

Every package is a normal desktop install: double-click, follow the wizard, and Stethos appears in the application menu / Start menu / Applications. No terminal, no Python, no model download — the inference engine and model weights are inside the package.

CPU builds (~330–500 MB)

Run everywhere. On Apple silicon macOS uses Metal (MPS) automatically.

Platform	File
Windows 10/11 x64	windows/Stethos_0.1.0_x64-setup.exe · windows/Stethos_0.1.0_x64_en-US.msi
macOS 11+ (Apple silicon)	macos/Stethos_0.1.0_aarch64.dmg
Debian / Ubuntu / Mint	linux/Stethos_0.1.0_amd64.deb
Fedora / openSUSE	linux/Stethos-0.1.0-1.x86_64.rpm
Arch / CachyOS / Manjaro	linux/Stethos-0.1.0-1-x86_64.pkg.tar.zst
Any Linux, no root	linux/Stethos_0.1.0_amd64.AppImage + linux/install-linux.sh

CUDA builds for Linux (~3 GB)

These bundle CUDA 13 and run inference on an NVIDIA GPU. They require a driver that supports CUDA 13.0 (580 or newer) and fall back to CPU otherwise. The filenames match the CPU build — install one or the other, not both.

Distro family	File
Debian / Ubuntu / Mint	cuda/linux/Stethos_0.1.0_amd64.deb
Fedora / openSUSE	cuda/linux/Stethos-0.1.0-1.x86_64.rpm
Arch / CachyOS / Manjaro	cuda/linux/Stethos-0.1.0-1-x86_64.pkg.tar.zst
Any Linux, no root	cuda/linux/Stethos_0.1.0_amd64.AppImage

Verify a download

Each platform folder has a SHA256SUMS.txt:

sha256sum -c SHA256SUMS.txt        # Linux
shasum -a 256 -c SHA256SUMS.txt    # macOS
certutil -hashfile <file> SHA256   # Windows

Check the GPU path

curl -s http://127.0.0.1:8765/health

"device": "cuda" with a non-null gpu_name means the CUDA build is active. The app's status bar shows the same information.

Notes

• macOS builds are ad-hoc signed. If Gatekeeper blocks the first launch, open System Settings → Privacy & Security and choose Open Anyway once.
• Windows installers are per-user and do not need an administrator account.
• Linux runtime dependencies: libwebkit2gtk-4.1, libgtk-3-0 (declared by the packages; the AppImage helper installs per-user under ~/.local).
• This is a research tool, not a medical device. Confirm findings with a clinician.