PicNote

AI-powered photo intelligence system. Monitors your iPhone photos (via iCloud sync), classifies informational vs casual images, extracts text/URLs/QR codes, and generates searchable Markdown notes automatically.

Zero friction capture — just take a screenshot or photo. PicNote does the rest.

How it works

iPhone → iCloud sync → Mac Photos Library → PicNote → Markdown notes

1. You take a photo or screenshot on your iPhone
2. iCloud syncs it to your Mac's Photos library
3. PicNote detects the new image via launchd (watches Photos.sqlite)
4. Classifies: informational (receipt, QR code, whiteboard, event flyer) or casual (selfie, scenery)
5. For informational images: extracts text (OCR), URLs, QR codes, contacts, amounts
6. Generates a structured Markdown note with AI analysis via Claude Code CLI
7. Saves to a configurable output directory (default: ~/Documents/PicNote/vault/)

Setup

Prerequisites

• macOS with iCloud Photos enabled
• Node.js 18+
• Claude Code CLI installed
• Xcode Command Line Tools (for Swift compiler)

Install

npm install -g picnote

# Or install from source
cd ~/Projects/picnote
npm install
npm run build

# Compile the Swift Vision OCR helper
swiftc -o swift/vision_ocr swift/vision_ocr.swift -framework AppKit -framework Vision

Configure

Edit config.yaml:

output_dir: ~/Documents/PicNote    # Where notes are saved
photos_library: ~/Pictures/Photos Library.photoslibrary

Run manually

# Process new photos
picnote

# Search notes
picnote --search "restaurant"

# View stats
picnote --stats

Auto-run on photo sync

# Install the launchd watcher (update ProgramArguments to use node + dist/main.js)
cp resources/com.picnote.watcher.plist ~/Library/LaunchAgents/
launchctl load ~/Library/LaunchAgents/com.picnote.watcher.plist

This watches your Photos.sqlite for changes and runs PicNote automatically within 30 seconds of a new photo syncing.

Architecture

| Component | Purpose | |-----------|---------| | src/watcher.py | Reads Photos.sqlite (strictly read-only via ?mode=ro) | | src/classifier.py | Informational vs casual classification (local heuristics + Claude CLI fallback) | | src/extractor.py | OCR via Apple Vision framework + regex extraction (URLs, phones, emails, amounts) | | src/analyzer.py | Deep analysis via Claude Code CLI — titles, summaries, tags | | src/note_generator.py | Generates Markdown notes with source traceability | | src/db.py | SQLite database with FTS5 full-text search index | | swift/vision_ocr | Compiled Swift binary for Apple Vision OCR + QR code detection |

Safety

• Photos are NEVER deleted or modified — strictly read-only access to your Photos library
• Photos.sqlite is opened with SQLite ?mode=ro (enforced at the VFS level)
• Path traversal protection on all file paths from the Photos database
• Sensitive content detection (passwords, SSN, bank info) skips cloud processing
• All Claude CLI calls use your existing Claude Code tokens — no separate API key needed

Tests

npm test

105 tests covering all modules, including safety tests that verify read-only guarantees.

License

Personal project.