StemDeck is an open-source, local audio stem separation tool designed for personal study and experimentation. Unlike cloud-based competitors like Moises or LALAL.AI, all processing occurs on the user's local machine using the Demucs htdemucs_6s model from Meta AI, which enables the extraction of up to six distinct stems: vocals, drums, bass, guitar, piano, and other. The application supports various input formats, including MP3, WAV, FLAC, OGG/Opus, MP4, and M4A, as well as YouTube URLs via yt-dlp. Technically, the software is built with a Python 3.12 backend using FastAPI and a Tauri v2 desktop shell (Rust/WebView2 or Rust/WKWebView). It features a DAW-style multitrack mixer with waveform rendering via , real-time VU meters, and audio analysis for BPM (via librosa), key/scale (Albrecht-Shanahan profiles), and LUFS (via pyloudnorm). Users can perform subset extraction to create custom mixes, which are summed using FFmpeg's amix filter. The tool provides hardware acceleration by automatically detecting compatible devices, including NVIDIA CUDA and Apple Silicon (MPS). By keeping all data local, StemDeck ensures privacy and eliminates subscription fees, requiring only an initial ~170 MB download for the Demucs model weights.