Research papers, distilled.
A beautiful, privacy-first research paper reader for iOS.
Browse, search, and summarize papers from arXiv — entirely on-device.
No login. No tracking. No analytics. No cloud. Just papers.
Papercut turns arXiv into a TikTok-style feed of research papers. One paper per screen. Swipe up to discover. Tap for AI-powered summaries. Everything happens on your device.
Swipe up through the latest and trending papers in your fields. Tap for AI magic — get TL;DRs, key findings, math explained, code walkthroughs, and more. Track research topics with persistent keyword-based feeds. Search anything across arXiv's 2M+ papers. Save for later with a double-tap.
Your reading habits, interests, and saved papers never leave your phone.
- Full-screen paper feed — Swipe vertically through papers like a social feed. Each card shows title, authors, categories, and an AI summary.
- Latest & Trending tabs — Latest papers sorted by submission date. Trending papers from the last 30 days sorted by relevance.
- New papers detection — Pull to refresh prepends new papers above your current position. A pill shows how many new papers arrived, with snap-back to resume reading where you left off.
- Scroll position persistence — Your reading position is saved per tab and restored across app restarts.
- Persistent bookmarks — Double-tap any paper to bookmark. Bookmarked papers survive cache cleanup and appear in the Saved tab.
- Custom topic feeds — Create keyword-based research topics (e.g., "RLHF", "diffusion models", "transformer efficiency") that act as persistent, auto-updating search feeds.
- Per-topic sorting — Sort each topic by relevance, newest, or oldest.
- Topic management — Add, remove, and browse topics from a dedicated tab.
- On-device AI summaries — 7 summary styles (TL;DR, Key Findings, Math Explained, Code Explained, Methodology, Implications, ELI5) generated entirely on-device using Apple Foundation Models. No API keys, no cloud processing.
- Smart summarization queue — Summaries are pre-generated in priority order: current paper first, then nearby papers, with intelligent priority demotion as you scroll.
- Powerful search — Full-text search across arXiv's 2M+ papers. Search by title, author, topic, or ArXiv ID.
- Search sorting — Sort results by relevance, newest first, or oldest first.
- Category filtering — Filter search results by arXiv categories.
- Explore topics — Quick-tap topic suggestions to kickstart your search.
- Daily digest — Configurable daily reminder at your chosen time.
- Topic update alerts — Get notified when new papers drop in your tracked topics.
- New feed items — Background checks for new papers in your followed categories.
- Smart onboarding — Notification setup explains why each type is useful before asking for permission.
- Category filtering — Follow specific arXiv categories (cs.AI, math.CO, physics.gen-ph, etc.) to curate your feed.
- Dark & light mode — Warm neutral palette with full dark mode support. Liquid Glass UI on iOS 26.
- Offline-ready — Papers and summaries are cached locally in SwiftData.
Papercut is built on a simple principle: your data stays on your device.
- No user accounts or login
- No analytics or telemetry
- No tracking of any kind
- No data sent to any server (except arXiv API for fetching papers)
- AI summaries generated on-device via Apple Foundation Models
- All preferences, bookmarks, and cached data stored locally
- iOS 26.0+
- Xcode 26+
- iPhone with Apple Silicon (for on-device AI summaries)
- Clone the repo:
git clone https://github.com/rajatady/Papercut.git
- Open
Papercut.xcodeprojin Xcode - Select your device or simulator
- Build and run
No API keys, no configuration, no .env files. It just works.
Papercut/
├── Models/ # SwiftData models (Paper, Summary, Topic, Author, Category)
├── Features/
│ ├── Feed/ # Main feed, full-screen cards, search, state machine
│ ├── Topics/ # Topic list, detail view, add topic sheet
│ ├── Onboarding/ # Category selection + notification setup
│ ├── Settings/ # App settings, category picker, summary styles
│ └── Feedback/ # In-app feedback with animated UI
├── Components/ # Reusable views (badges, chips, pills, pickers)
├── Services/
│ ├── ArXiv/ # ArXiv API client, XML parser, endpoints
│ ├── Summarization/ # On-device AI with @Generable schemas
│ ├── Notifications/ # Local notification scheduling + types
│ ├── Background/ # BGAppRefreshTask checkers for topics + feed
│ ├── Repositories/ # Data layer (PaperRepository, TopicRepository, PreferencesStore)
│ └── Storage/ # Local summary cache
└── Theme/ # Design system (colors, typography, spacing, animations)
Key design decisions:
- Pure state machine — Feed uses a deterministic
TabStateMachinewith typed events and side effects. Each tab owns independent state. 97+ tests cover transitions and invariants. - SwiftData for persistence — papers, summaries, topics, and bookmarks
@Observableview models with@MainActorisolation@Generablestructured output — AI summaries use Apple's constrained decoding for guaranteed clean output (no markdown, no emojis, controlled length)- Priority-based summarization queue — critical/high/medium/low priorities with automatic demotion
- Background task infrastructure —
BGAppRefreshTaskfor topic and feed update checks with per-topic notifications
372 tests covering:
- State machine transitions and invariants
- ViewModel integration tests
- Repository CRUD operations
- Topic management
- Notification preferences and backward compatibility
- Background task configuration
xcodebuild test -scheme Papercut -destination "platform=iOS Simulator,name=iPhone 17 Pro" -parallel-testing-enabled NOContributions are welcome! Please feel free to submit issues and pull requests.
MIT License. See LICENSE for details.
Built with SwiftUI, SwiftData, and Apple Foundation Models.