Compatibility
Support matrix for choosing a Toposync installation path.
Systems
| System | Status | Recommended path |
|---|---|---|
| Linux | Supported | Python, Docker CPU, Docker CUDA, or processing server |
| macOS | Supported for CPU | Python or processing server |
| Windows | Supported | Python or processing server as a Windows service |
| Home Assistant OS | Supported through the add-on | Home Assistant add-on |
Architectures
| Architecture | Status | Notes |
|---|---|---|
amd64 / x86_64 | Supported | Main path for Linux, Windows, Docker, and HAOS |
aarch64 / arm64 | Supported for CPU | Supported path for Docker CPU and Home Assistant OS |
| Apple Silicon | Supported for CPU | Use the Python installation on macOS |
| Intel macOS | Supported for CPU | Use the Python installation on macOS |
armv7, armhf, i386 | Outside the support target | Use a 64-bit system |
Installation scenarios
| Scenario | Status | Notes |
|---|---|---|
| Python on Linux and macOS | Supported | Recommended with Python 3.12 |
| Python on Windows | Supported | Recommended with Python 3.12 |
| Docker CPU | Supported | Validated for amd64 and arm64 |
| Docker CUDA | Linux + NVIDIA | Requires NVIDIA driver and NVIDIA Container Toolkit |
| Home Assistant add-on | Supported | CPU-only on amd64 and aarch64 |
| Processing server on Linux and macOS | Supported | CPU on macOS; CPU or CUDA on Linux |
| Processing server on Windows | Supported | CPU, DirectML, or native CUDA |
| Processing server on Docker | Supported | CPU; CUDA when the local image was built with the CUDA target |
GPU
| Acceleration | Status | Use when |
|---|---|---|
| CPU | Default | First install, light usage, and broad compatibility |
| CUDA on Linux | Supported | Linux host with an NVIDIA GPU |
| CUDA in Docker | Supported on Linux | Linux host with an NVIDIA GPU and NVIDIA Container Toolkit |
| Native CUDA on Windows | Supported as a Python bundle | Windows machine with compatible NVIDIA driver and runtime |
| DirectML on Windows | Supported | Windows GPU compatible with DirectML |
| CUDA in the Home Assistant add-on | Outside the current scope | Use an external processing server if you need GPU acceleration |
Raspberry Pi and HAOS
| Environment | Status | Recommendation |
|---|---|---|
| Raspberry Pi with 64-bit HAOS | Supported through the add-on | Use aarch64 |
| Raspberry Pi 5 8 GB + NVMe | Practical reference | Better baseline for modern usage |
| Raspberry Pi 4 | Best-effort | Suitable for light usage and compatibility testing |
| SD card storage | Best-effort | Avoid it for many cameras or write-heavy workloads |
| Heavy vision/OpenCV on ARM CPU | Limited | Delegate to a remote processing server |
Practical rule
Start with CPU. Add GPU acceleration or a processing server only when you hit a real bottleneck in vision, OpenCV, multiple cameras, or ONNX inference.