## Linux

| Requirement | Minimum |
| --- | --- |
| Architecture | x86_64 or ARM64 |
| GPU | CUDA 11.8+ compatible |
| RAM | 512 MB |

## Windows

| Requirement | Minimum |
| --- | --- |
| Architecture | x86_64 or ARM64 |
| OS | Windows 10 (build 10.0.19041+) |
| GPU | DirectX 12 compatible |
| RAM | 512 MB |

## Android

| Requirement | Minimum |
| --- | --- |
| Architecture | ARM64 |
| API Level | 26 (Android 8.0+) |
| GPU | OpenGL ES 3.1+ |
| RAM | 512 MB |

## iOS

| Requirement | Minimum |
| --- | --- |
| Architecture | ARM64 |
| OS | iOS 17.6+ |
| GPU | Metal-compatible |
| RAM | 512 MB |
| Neural Engine | Optional, improves performance |

## macOS

| Requirement | Minimum |
| --- | --- |
| Architecture | ARM (Apple Silicon) or Intel x86_64 |
| OS | macOS 14.6+ |
| GPU | Metal-compatible |
| RAM | 512 MB |
| Neural Engine | Optional (Apple Silicon only) |

## General notes

- GPU acceleration is used by default. Fall back to CPU with the `useCPUBackendOnly` flag if needed.
- RAM requirements are per model. Running multiple models simultaneously requires more.
- Model files (`.crypt`) are provided by AudioShake and tied to your Client ID and Secret.