HN: AI File Sorter auto-organizes files using local AI (Windows, macOS binaries) (github.com)

AI File Sorter 0.9.7 is a cross‑platform desktop app (GTK UI) that automates file organization using both local and remote LLMs. It classifies files and optionally subcategories based on filenames/extensions, shows a preview so you can edit assignments, then creates folders and moves files. The release tightens categorization consistency via a taxonomy, adds file path context to LLM prompts, improves logging and CUDA detection, and brings CUDA 13 compatibility—so local inference (llama.cpp with LLaMa/Mistral) can be accelerated on Nvidia GPUs while falling back to OpenBLAS/CPU. Remote sorting via ChatGPT o4‑mini is supported with encrypted API key storage and optional obfuscation. For AI/ML practitioners this is notable as a practical, offline‑friendly application of LLMs to a common productivity task: local model support avoids API costs and data exposure while retaining a remote option for higher quality or convenience. Technical implications include reliance on llama.cpp builds (OpenCL not enabled by default in the build script), optional CUDA acceleration, local DB caching to reduce repeated remote calls, and detailed platform build flows (Windows/MSYS2+Visual Studio, macOS/Homebrew+Xcode, and multiple Linux distro instructions). The project is aimed at users comfortable compiling dependencies, and includes NAS/remote directory sorting, secure API key handling, and improved stability/logging for production usage.