| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| ExtremeCleaner.exe | 2026-11-10 | 50.3 MB | |
| 1.3.1 - Scribe speech models get per-model rows source code.tar.gz | 2026-11-10 | 2.2 MB | |
| 1.3.1 - Scribe speech models get per-model rows source code.zip | 2026-11-10 | 2.3 MB | |
| README.md | 2026-11-10 | 1.2 kB | |
| Totals: 4 Items | 54.8 MB | 3 | |
Adds ExtremeCleaner's first sibling-app rule: Scribe speech-recognition models.
On the User data tab a new AI-models row scans %LOCALAPPDATA%\Scribe\models (developer-confirmed layout models/<language>/<model>) and lists what is actually installed, one row per model - vosk language packs, punctuation models, whatever Scribe has downloaded. Nothing installed = no row at all.
The app never deletes models itself: rows are review-only, a click opens the model folder in Explorer. Models are user data - delete one by hand if you no longer use that language; Scribe re-downloads it on next use.
Under the hood: the folder-discovery mechanism already used by the OneDrive hub gained a depth parameter (discover_depth), so nested layouts like Scribe's models/<lang>/<model> become one row per leaf - 47 files of a vosk model collapse into a single understandable row instead of 10 subfolder rows.
Rules: 297 (+1). Smoke-tested: scan, totals, rendered rows all green. SourceForge: Files -> 1.3.1.
