FocusSort is built native, not wrapped. Metal for rendering, Core Data for indexing, Neural Engine for AI. No compromises in the stack.
Tested on real libraries. Not synthetic benchmarks.
FocusSort is built entirely in Swift and SwiftUI. Here's what that means for performance.
Every thumbnail, grid view, and zoom transition goes through Metal — Apple's low-level GPU framework. This lets us saturate the GPU directly without UIKit abstraction overhead. Result: consistent 60fps even with thousands of items on screen.
All AI models are compiled for CoreML and run on the Apple Neural Engine when available (M1 and later). This frees up the CPU for rendering and keeps inference fast and power-efficient — no GPU compute cost during AI processing.
Import, thumbnail generation, and AI analysis all happen in background threads that yield to the UI. You can browse your library immediately after import starts — processing happens silently without blocking you.
Thumbnails are cached at multiple resolutions and loaded from disk in parallel. The predictive cache pre-loads thumbnails for items about to scroll into view, eliminating the blank-frame flicker common in other photo managers.