Pre-alpha · 2026
tidyup
A local-first file organizer that classifies by content, proposes a better structure, and keeps every operation reviewable and reversible.
On-device by default
Most “smart” file organizers send sensitive documents to a remote model. tidyup takes the opposite approach: it runs entirely on-device by default, ships with no HTTP client, and asks before changing anything.
The system recursively scans a directory, extracts useful signals from file contents, and classifies each item through a three-tier cascade. Fast heuristics handle obvious cases, local embeddings handle semantic matching, and an optional model can be enabled for low-confidence results.
Design constraints
- Every move is presented as a proposal before it happens.
- Related files are treated as an atomic bundle instead of being scattered.
- Originals are shelved before changes, making operations reversible.
- The core is separated from its CLI and desktop interfaces through a hexagonal crate layout.
The project is under active development and intentionally conservative around real user data.