- Added document content analysis (text LLM) with rename-only/document-only options and optional creation-date suffixes for categories. Supported document formats include PDF, DOCX, XLSX, PPTX, ODT, ODS, and ODP (plus common text formats).
- Local 3B model download now defaults to Q4 for better GPU compatibility. The legacy Local 3B Q8 is still selectable when an existing download is found.
- Improved the LLM selection dialog latency.
- Added custom API endpoints to the Select LLM dialog. Custom endpoints accept base URLs or full /chat/completions endpoints, with optional API keys for local servers.
- LLM-derived categorizations and rename suggestions are now saved as you go, so progress isn't lost if the app closes unexpectedly.
- Image analysis now falls back (with a user prompt) to CPU if the GPU has insufficient available memory.
- Review dialog now lets you select highlighted rows and bulk edit their categories.
- Review dialog is now scrollable on smaller screens so action buttons stay visible.
- Improved subcategory consistency by merging labels that only differ by generic suffixes (e.g., “files”).
- Added a system compatibility check (benchmarking) to determine the most suitable LLM for your system.
- Added Korean as an interface language.
- macOS builds now include variant `make` targets for Apple Silicon (M1 / M2-M3) and Intel outputs, plus improved arch-aware llama.cpp builds.
- UI, stability, persistence, and usability improvements.
PyAppExec is a lightweight cross-platform bootstrapper/launcher for distributing Python desktop applications without freezing them into large binaries with PyInstaller / Nuitka / cx_Freeze.
Instead of bundling Python and all dependencies into a single executable, PyAppExec automatically sets up a virtual environment on first launch, installs dependencies (and optional third-party tools like ffmpeg), and then runs the entry script directly. This avoids antivirus false positives and results in much smaller downloads.
It includes both an installer and a bootstrapper (CLI and GUI variants). Currently Windows-only binaries are available; macOS and Linux builds are in progress.
Use cases include distributing Python tools to non-technical end users, apps that need external binaries, and workflows where update size matters.
Sharing an update on AI File Sorter, a cross-platform desktop app for cleaning up messy folders (Downloads, NAS, external drives) using local LLMs for automated file categorization.
Runs fully offline, no telemetry, no cloud.
What’s new in v1.3:
- Add your own Local LLM (custom .gguf models in the Select LLM dialog).
- Two categorization modes - More Refined vs More Consistent.
- Optional Whitelists to limit allowed category names.
It helps tidy up cluttered folders like Downloads or external/NAS drives by automatically categorizing files based on their names, extensions, directory context, and taxonomy.
It uses a taxonomy-based system, so the more files you sort, the more consistent and accurate the categories become over time. It essentially builds up a smarter internal reference for your file types and naming patterns. Also, file content-based sorting for some file types is coming up as well.
The app features an intuitive, modern Qt-based interface. It runs LLMs locally and doesn’t require an internet connection unless you choose to use the remote model. The local models currently supported are LLaMa 3B and Mistral 7B.
The app is open source and Metal-optimized for macOS. Supports CUDA and CPU on Windows.
AI File Sorter is an intelligent, AI-powered file management tool that helps you organize files effortlessly. Using cutting-edge machine learning, it categorizes files automatically based solely on their names and extensions—ensuring your privacy is maintained. Only the file names are sent to the LLM (ChatGPT), with no other data shared, making it a secure and efficient solution for file organization.
The app is now open-sourced. It is also cross-platform, and a pre-compiled installer is available for Windows.
I’ve built a free tool called AI File Sorter. It uses LLMs to intelligently categorize and organize files and directories, which I think might be handy for many people. Well, let’s just say it can finally bring some order to the chaos of your Desktop or Downloads folder. But it will also handy for those who've got a lot of data in general.
It’s fast and reliable, written in C++ (current version: 0.8.0).
You "can ask for the file" by searching for it, but it will necessarily take some time. Another approach would be to maintain and always update a database of files with locations. But it's still a good idea to keep files organized, such that each item has its own place.
New in v1.7.0:
- redesigned the progress dialog for clearer progress tracking
- added metadata-based rename suggestions for audio/video files
- improved reliability and stability