1-Click Private Offline AI Translation on Apple Silicon Mac
Whether you work with confidential legal contracts, clinical notes, NDA source code, or travel offline, Subtle runs neural machine translation 100% on your Mac—with zero terminal setup and zero background RAM bloat.
How Subtle’s Offline Engine Works
- Sandboxed Apple Silicon Metal Helper: Executes pure SHA-256-verified
.ggufweights inside an isolated App Sandbox Metal helper over anonymous pipes (zero open network ports). - Standard (4B) & High Precision (12B) Tiers: Pick fast everyday offline lookups or deeper literary/idiomatic nuance in Settings → Offline.
- Automatic 120-Second & Memory-Pressure Eviction: Unlike background LLM servers that hold gigabytes of unified memory all day, Subtle automatically unloads model weights after 2 minutes of inactivity (or immediately under macOS memory pressure), returning to ≤ 35 MB idle RAM.