Your Downloads folder has 400+ files you've been "meaning to sort" since 2023. SlothSort triages them for you — with TabPFN, an open-source tabular foundation model running locally on your CPU. The 24-file demo folder below scores in a single ~9-second forward pass.
🔒 100% local & private: SlothSort sees only file metadata (name, size, dates) — never contents. Nothing leaves your machine; it works fully offline, set up once, never pay.
⚠️ PRECOMPUTED + SIMULATED demo (baked ): predictions below were computed by on an invented ~/Downloads set — no real files were scanned. The live Flask app (app.py) scores the same set on request.
Simulated ~/Downloads — triaged in seconds
–keep
–archive
–delete
–reclaimable
File
Verdict ⓘ
Why ⓘ
Do it
How it works
Metadata features — size, age, days since opened, file-type category, filename clutter signals (final, copy, v2…), hidden/temp flags. Contents never touched.
TabPFN (open-source tabular foundation model, local CPU, no API key) classifies each file KEEP / ARCHIVE / DELETE from 6,000 simulated file records.
Act — one click archives or deletes, always with undo. The real script moves files on your machine; this demo only simulates.
Built Oct 20, 2026 for the DEV Hacktoberfest AI Challenge (Week 3: Slow and Steady). MIT licensed. Repo: github.com/arindewangan/slothsort