> ## Documentation Index > Fetch the complete documentation index at: https://developer.audioshake.ai/llms.txt > Use this file to discover all available pages before exploring further. # Overview > Run audio source separation on-device with the AudioShake Local Inference SDK The Local Inference SDK runs AudioShake's separation models directly on-device — no network calls, no cloud processing. It ships as a native library for Windows, Android, iOS, and macOS. ## Use cases * **Real-time processing** — separate stems live from a microphone or audio stream * **Offline applications** — process audio without an internet connection * **Low-latency pipelines** — integrate into DAWs, game engines, or embedded systems * **Privacy-sensitive workflows** — keep audio on-device ## Supported platforms | Platform | Architecture | GPU backend | CPU supported | | -------- | -------------- | -------------------- | ------------- | | Linux | x86\_64, ARM64 | CUDA 11.8+ | Yes | | Windows | x86\_64, ARM64 | DirectX 12, CUDA | Yes | | Android | ARM64 | OpenGL ES 3.1+ | Yes | | iOS | ARM64 | Metal, Neural Engine | Yes | | macOS | ARM, x86\_64 | Metal, Neural Engine | Yes | Demo applications are available for each platform. ## How it works The SDK loads an encrypted `.crypt` model file provided by AudioShake. You pass audio buffers in, and separated stem buffers come out. Processing runs on CPU or GPU depending on platform and configuration. | Interface | Best for | | ---------------------- | -------------------------------------------------------------- | | `SourceSeparationTask` | File-to-file separation with progress tracking | | `AudioShakeSeparator` | Stream and buffer-based processing for real-time or custom I/O | ## Get access The SDK requires a **Client ID** and **Client Secret** for authentication and model decryption. Contact [info@audioshake.ai](mailto:info@audioshake.ai) to get started with the Local Inference SDK.