Sift

Sift solves the "save and forget" problem: you bookmark a post meaning to return to it, and never do. Share a post or link into Sift through the iOS share sheet — no account, no setup — and it pulls in the caption and thumbnail, reads the content with on-device AI, files it under the right category, and brings it back with an end-of-day summary or a reminder you set on any item.

Classification runs on Apple's Foundation Models on-device by default and falls back to cloud Gemini on older hardware — both behind a single LanguageModel protocol that uses guided generation into typed Swift structs. In-app search is semantic rather than keyword-based, built on NLContextualEmbedding vectors ranked by cosine similarity, so it works offline and nothing you type leaves the device.

The app is built in Swift 6 with strict concurrency and a modular SwiftKit architecture (SiftCore / SiftStorage / SiftEnrichment / SiftFeatures / DesignSystem) whose module boundaries are enforced in CI. Storage is offline-first — a SwiftData cache synced to Firestore behind a ContentRepository protocol — with a capture Share Extension that durably records to an App Group inbox in milliseconds and background enrichment driven by BGTaskScheduler. Auth starts anonymous and upgrades to Sign in with Apple or Google, hardened with Firebase App Check. Shipped to the App Store with a Cloud Functions backend and a Swift Testing suite.