Check the supported edition and operating system path, then size the host for the models and backends you plan to run.
Software and operating system requirements
| Edition and platform | Current installation path |
|---|---|
| SoAI Core on Linux | The published Linux archive supports 64-bit x64 and ARM64 hosts running Debian 12 or 13 or Ubuntu 24.x or 26.x. |
| SoAI Core on macOS | The published complete archive supports Intel and Apple silicon and provides an in-application update path. |
| SoAI Core on Windows | The published installer supports 64-bit Windows 10 or later on x64. There is no Windows ARM64 installer. |
| SoAI OS | The appliance edition uses a UEFI Debian 13 KDE image for AMD64 and ARM64 hosts. |
Platform support does not guarantee a particular inference backend or GPU feature. Check the backend's own requirements and the model's memory needs before deploying it.
Operating system tiers
Minimum: Debian 12 or 13, Ubuntu 24.x or 26.x, Windows 10 or 11, Windows Server 2016, 2019, 2022, or 2025, or macOS. Recommended: Debian 13, Ubuntu 26.x, Windows 11, Windows Server 2025, or macOS.
The operating system tiers apply to SoAI Core. SoAI OS uses the appliance image listed above. The selected backend may have additional operating system, driver, and runtime requirements.
The standard Linux installer requires root or sudo access to install
system packages and write to /opt/soai. Automatic service startup requires
systemd. The macOS installer requires
launchctl and plutil when automatic startup is selected. A first installation needs a
temporary internet connection to obtain the release, managed runtime dependencies, and required artifacts.
See Install SoAI Core for the installation and privilege choices, and
Offline installation and operation for preparing an offline host.
Hardware requirements
SoAI works on CPU-only machines and on machines with one or more GPUs. A GPU is optional, but it is needed for reasonable local inference speeds.
| Resource | Minimum | Recommended |
|---|---|---|
| CPU | Intel Core i5-2500 or AMD FX-6100 | Intel Core i7-6700 or AMD Ryzen 5 1600 |
| RAM | 4 GB DDR3 | 16 GB DDR4 |
| Storage | 128 GB SATA SSD (SoAI occupies about 5 GB when freshly installed) | 512 GB NVMe SSD (SoAI occupies about 5 GB when freshly installed) |
| GPU | Optional compatible NVIDIA Pascal-generation GeForce GTX 10-series, Quadro P-series, or Tesla P-series cards; or an AMD or Intel GPU supported by the selected backend's ROCm, Vulkan, XPU, SYCL, OpenVINO, or OpenCL runtime | Compatible NVIDIA Ampere-generation GeForce RTX 30-series, RTX A-series, or Ampere data-center cards; AMD Radeon RX 6000/7000-series or Instinct MI200/MI300-series GPUs; or Intel Arc A-series/B-series or Data Center GPU Max cards |
RAM and VRAM needs depend on the selected model, quantization, context length, plugin, and concurrency rather than on SoAI alone. Larger models and concurrent requests need more memory. Model files, plugin backends, databases, user files, backups, and working space also need storage beyond the application itself.
SoAI can run from a hard disk drive (HDD), SD card, or USB drive, but those storage types are not recommended and do not meet the SSD-based minimum or recommended storage tiers. Their lower throughput, higher latency, and removable-media failure risks can affect model loading, database activity, indexing, updates, backups, and concurrent file operations. Use an SSD for a normal installation and durable data paths.
Use an enterprise SSD for heavy or enterprise workloads. For maximum persistent-storage performance, an Intel Optane drive is the best option where available. A RAM drive can be used for maximum performance, but it is not recommended.