Tagme

Tagme is an AI file auto-labeler for macOS that turns messy screenshots and downloads into searchable, context-rich filenames.

It combines Tesseract OCR and Ollama vision (llava:7b) to identify:

• source app/site (floom, github, linkedin)
• screen intent (token-setup, dashboard, signin)
• task context (mcp-config, api-token, agent-auth)

Why Tagme

• Automatic desktop cleanup without deleting files
• Better screenshot search and retrieval
• Works locally with private on-device inference
• Optimized for founders, operators, and heavy screenshot workflows

Example output

Input:

• Screenshot 2026-04-28 at 20.05.00.png

Output:

• 2026-04-28__screenshot-floom-token-setup-mcp-config__screenshot-2026-04-28-at-20-05.png

Features

• Watches Desktop, Downloads, and Documents
• OCR text extraction via tesseract
• Vision context extraction via ollama + llava:7b
• Bucketed labels: source + screen_type + task_topic
• SQLite state tracking and log history
• launchd automation for background scans

Install (macOS)

1. Install dependencies:

brew install tesseract imagemagick ollama

2. Pull model:

ollama pull llava:7b

3. Install Python package locally:

pip install .

4. Copy config:

mkdir -p ~/.config/file-labeler
cp config/config.example.json ~/.config/file-labeler/config.json

5. Install launch agent:

cp launchd/com.federico.filelabeler.plist ~/Library/LaunchAgents/
launchctl unload ~/Library/LaunchAgents/com.federico.filelabeler.plist >/dev/null 2>&1 || true
launchctl load ~/Library/LaunchAgents/com.federico.filelabeler.plist

CLI

Run one scan:

tagme

Config

Config file:

• ~/.config/file-labeler/config.json

Main keys:

• watch_dirs
• enabled_since
• rename
• ollama.model
• ocr.enabled

Logs and DB

• Log: ~/.local/state/file-labeler/labeler.log
• State DB: ~/.local/state/file-labeler/labels.db

npm wrapper

This repo includes @floomhq/tagme for JS users. It wraps the Python CLI:

npx @floomhq/tagme

PyPI

Package name:

• tagme

License

MIT