Floxtop is a native macOS app that organizes your files automatically using an on-device AI model, with no data leaving your machine. It focuses on privacy, speed, and usability.

Background:

Manual file organization is tedious, and traditional tools rely on filenames or metadata, missing file meaning. Floxtop uses Sentence Transformers and computer vision including OCR and transformer-based embeddings to analyze content across PDFs, images, Office files, EPUBs, CSV, Markdown and more (30+ formats supported).

Key features:

- Entirely offline AI inference, optimized for Apple Silicon, ensuring privacy and near-instant processing.

- Multi-format content understanding via OCR and semantic embeddings to group files by meaning including images, Microsoft Office, PDFs, EPUBs, CSV, Markdown and more (+30 file types supported).

- Customizable classification: define categories to define file destinations.

- Native macOS integration with Finder extensions and Quick Look previews.

Performance:

Floxtop categorizes files in under a second. Built in Swift with ONNX, the app is compact (~120 MB) with a powerful 94 MB model.

I launched an early beta, Fallinorg (https://news.ycombinator.com/item?id=44932375), reaching 103 customers and gathering valuable feedback. Floxtop is the improved iteration with broader format support, compact binary size, and better integration.

Try the free version: https://pub-a80e0e01ccee4148a4f4c43d8d4eeda2.r2.dev/Floxtop-...

I’d appreciate feedback on accuracy, performance, and feature ideas.

Thanks for checking it out!