Scan, extract, file — offline on your own device. 2.1.3 finally reads metadata in your language instead of English.
Archi is the Paperless-NGX client for iPhone, iPad and Mac that runs AI and OCR entirely on your device — no cloud subscription, no data sharing, zero privacy compromises.
Note: AI metadata extraction requires a one-time download of the Gemma 4 model (approx. 2.4 GB) after installation.
■ WHAT ARCHI DOES
Scan a document, recognize text, extract metadata — all in seconds, all locally. The document then lands neatly categorized in your self-hosted Paperless-NGX server.
■ ON YOUR DEVICE, NOT IN THE CLOUD
• OCR with Apple Vision — fast, offline, free
• AI metadata extraction with Google's Gemma 4 — runs directly on your iPhone, no server needed
• Automatic title, tag and correspondent suggestions based on content
• Import PDFs and images from other apps via the Share sheet
■ PAPERLESS-NGX INTEGRATION
• Browse and manage your entire document archive
• Full-text search across all documents
• Filter by tags, document type and correspondent
• Offline queue — scans saved locally until you're back online
■ REAL-WORLD USE CASES
Tax Season: Scan every receipt, invoice and bill as it arrives. Archi automatically recognizes the date, vendor and amount — and suggests the right tags. Come April, your IRS paperwork is already done.
Home Office & Business: Incoming invoices, delivery notes, contracts — Archi reads the document and proposes correspondent, document type and filing date without you lifting a finger.
Medical & Insurance: Digitize EOBs, lab results, insurance policies and medical records securely. Your health data never leaves your device.
No Wi-Fi? No Problem: Scan documents in the subway, in a basement archive or while travelling. Archi queues everything locally and syncs automatically when you're back online — unlike server-dependent alternatives.
Homelab & Self-Hosting: Running Paperless-NGX on a Synology NAS, Proxmox, Docker or a Raspberry Pi? Archi was built for exactly your setup. No SaaS, no monthly fees, no vendor lock-in.
■ PRIVACY FIRST
One-time purchase. Zero subscriptions. Zero cloud. Zero data collection.
Your documents stay on your device and your server — and nowhere else.
All AI processing, all OCR, all metadata extraction runs exclusively on your iPhone. Archi never transmits document content to any external server — not Apple's, not Google's, not ours.
■ TECHNICAL DETAILS
• AI model: Gemma 4 E2B INT4 (Google open weights) — runs fully on-device via LiteRT, Metal GPU accelerated
• OCR: Apple Vision Framework — free, offline, privacy-preserving
• Backend: Paperless-NGX (self-hosted, URL and token fully configurable)
• Offline queue: SQLite-based, zero data loss when offline
• On-device AI: iPhone 15 Pro or newer, or an iPad with 6 GB RAM (M-series). Scanning, uploading and your archive work on all supported devices — even without AI.
• Runs on iPhone, iPad and Apple silicon Mac.
Archi requires a self-hosted Paperless-NGX server.
Before purchasing it’d be helpful to know that: a 2.4GB download is required; have to use SSL/TLS even on local network; ‘suchen’ & ‘scannen’ even if English is chosen language; dates are day-month-year not month-day-year so you’ll correct every scan (in US); auto scan crops aggressively, so have to use manual; crashes when processing more than one scan (iPhone 16 Pro, iOS 26.5); support is only if you have a GitHub account. Disappointed.
Developer Response
Thank you for the detailed feedback — these are fair points and several are already being worked on.HTTP/local network: a toggle for this exists in Settings → Security (and now also in the server setup screen as of v1.4.1 — sorry it wasn't visible during your setup).Date format: US date order (MM/DD) is on my list — I'll add it as a locale option.Model download size: I'll add this to the App Store description so it's clear upfront.Crashes with multiple scans: known issue, being fixed.Support without GitHub: noted — I'll add an alternative contact option.I'm sorry the experience fell short.
New in 2.1.3
Fixes
• The "automatic" metadata language now follows your device language instead of falling back to English. This affected Dutch, Polish, Ukrainian and Brazilian Portuguese — despite specially tuned prompts, those users ended up with English titles. If you set the language manually, nothing changes for you.
Improvements
• Tag suggestions now follow what you type: typing narrows the list, and the tags you use most appear first.
Version 2.1.3
The developer, Arne Stenmanns, indicated that the app’s privacy practices may include handling of data as described below. For more information, see the developer’s privacy policy .
Data Not Collected
The developer does not collect any data from this app.
Privacy practices may vary, for example, based on the features you use or your age. Learn More
Accessibility
The developer has not yet indicated which accessibility features this app supports. Learn More
Information
Seller
Arne Stenmanns
Size
59.3 MB
Category
Productivity
Compatibility
Requires iOS 26.0 or later.
iPhone Requires iOS 26.0 or later.
iPad Requires iPadOS 26.0 or later.
Mac Requires macOS 26.0 or later and a Mac with Apple M1 chip or later.
