Clip Learner — iOS

A native SwiftUI app for learning English from YouTube clips. It's a thin client for the existing clip-learner SvelteKit backend — all heavy work (yt-dlp download, Whisper transcription, LLM analysis) runs on the server. The app authenticates with the same accounts and shows the episodes / notebook you generate on the web.

Run

cd cliplearn
open cliplearn.xcodeproj   # build & run (⌘R). Requires Xcode 26+, iOS 26 SDK.

YouTubePlayerKit is vendored as a local Swift package (../YouTubePlayerKit-main, committed in this repo), so it builds straight after clone — no package fetch needed.

Backend connection

The app talks to the backend over JSON. The base URL lives in cliplearn/cliplearn/APIClient.swift:

enum APIConfig {
    static let localDev    = URL(string: "http://localhost:5174")!   // npm run dev
    static let production  = URL(string: "http://43.134.87.27")!     // the VPS
    static let baseURL = production   // ← flip to .localDev for local backend work
}

• Session uses the backend's clip_session cookie via URLSession's shared cookie storage (auto-persisted). Login is on-demand (a sheet), not a launch wall.
• The VPS is currently plain HTTP, so the cookie is secure:false server-side and Info.plist has ATS exceptions for localhost + 43.134.87.27. Once the VPS is on HTTPS, revert those and drop the exceptions.

Structure (cliplearn/cliplearn/)

File	Role
APIClient.swift	Typed wrapper over URLSession; episodes, detail, explain, notebook, auth, add/delete.
AuthStore.swift / AuthView.swift	Login state + on-demand login sheet.
Episode.swift / Models.swift	Codable models mirroring the backend.
RootView.swift	iOS 26 Liquid-Glass TabView: Home / Notebook / Profile.
HomeView.swift	Episode feed: list, swipe-delete, long-press category, filter chips, + add clip with status polling.
StudyLoaderView.swift / StudyView.swift	Loads transcript, then the core study screen: YouTube player + immersive scrollable caption (active-word highlight), tap-word lookup popover, custom transport bar, save-line.
WordLookupSheet.swift	Word/phrase explanation popover (/api/explain, LLM).
NotebookView.swift	Synced vocabulary with search + category filter.

Backend endpoints used

GET /api/episodes, GET|PATCH /api/episodes/[id], POST /api/auth, POST /api/explain, GET|POST|DELETE /api/notebook, POST|DELETE /api/process, GET /api/episode/[id]/status.

Notes / TODO

• HTTPS on the VPS (domain + Let's Encrypt) → removes the cleartext-cookie risk.
• Background audio mode (would need the backend to retain audio + AVPlayer).
• Word-level highlight is approximate (per-line timestamps only); precise sync needs Whisper word timestamps stored server-side.
• yt-dlp on the VPS uses a synced cookies.txt to dodge YouTube's bot check.