Pluck

AI-powered desktop tool that extracts product information from any shopping page — no API key required.

• Local mode (free, on-device) — PaddleOCR + HTML parsing extracts name, price, discount, and product URLs without any cloud calls.
• Cloud mode (optional) — choose Claude, GPT-4o, Gemini, or Ollama (LLaVA) for higher-accuracy vision extraction.
• Custom instructions — type "extract size, color, material" and the model returns those fields under extra for every product.
• Modern desktop GUI — sidebar engine selector, sleek monogram brand, paste-friendly URL input.
• Subscription tiers (Free / Pro / Team) with HMAC-signed license keys.

Install

cd shopping-extractor
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
.venv/bin/playwright install chromium

Cloud engines are optional. If you're using the default local engine, you can skip API keys entirely. To use a cloud engine, set the matching env var in a .env file:

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
GEMINI_API_KEY=...

Run

.venv/bin/python main.py --gui                                       # GUI
.venv/bin/python main.py --url "https://books.toscrape.com/"         # CLI, local engine (free)
.venv/bin/python main.py --url "..." --engine claude                 # cloud Vision
.venv/bin/python main.py --url "..." --engine gemini                 # Google free tier
.venv/bin/python main.py --url "..." --engine ollama --model llava   # fully offline
.venv/bin/python main.py --url "..." -i "extract size and color"     # custom fields

Flag	Description
--url	Page to extract
--engine	local / claude / openai / gemini / ollama (default local)
--model	Override engine's default model
--api-key	API key for cloud engines (or use the env var)
--max-products N	Limit number of products (default 20)
--output DIR	Output folder (default ./output)
--no-headless	Show browser window
--instructions / -i	Free-form instruction for the LLM
--gui	Launch GUI
--mint-key PRO|TEAM	Generate a demo license key
--activate KEY	Activate a license key
--plan-status	Show plan, expiry, and today's usage

Engines

Engine	API key	Cost	Accuracy	Notes
local	❌	Free	★★★	PaddleOCR/EasyOCR/Tesseract — first install ~300 MB
claude	✅	Paid	★★★★★	Best on cluttered cards, Korean/Japanese
openai	✅	Paid	★★★★	GPT-4o-mini is fast and cheap
gemini	✅	Free tier	★★★★	Generous free quota
ollama	❌	Free	★★★	Fully offline, requires ~4 GB LLaVA model

Output

output/
├── images/
│   ├── product_001.png
│   └── ...
├── page_full.png
├── results.json
└── results.csv

results.json example:

{
  "url": "...",
  "engine": "local",
  "custom_instructions": "extract size and color",
  "products": [
    {
      "id": 1,
      "name": "...",
      "price": "$29.99",
      "discount_rate": "30%",
      "image_path": "output/images/product_001.png",
      "product_url": "...",
      "extra": { "size": "L", "color": "black" }
    }
  ]
}

Subscription

Plan	Price	Daily extractions	Max products / run	Custom instructions	Batch URLs
Free	$0	5	10	—	—
Pro	$9.99 / mo	unlimited	50	✓	—
Team	$29.99 / mo	unlimited	200	✓	✓

.venv/bin/python main.py --mint-key PRO --days 365
.venv/bin/python main.py --activate PRO-2027...-...
.venv/bin/python main.py --plan-status

In the GUI, click the plan badge in the sidebar to open the subscription dialog.

Modules

File	Role
main.py	CLI entry + license utilities
gui.py	CustomTkinter app + subscription dialog
subscription.py	Plans, license signing, usage tracking
scraper.py	Playwright page load + product card cropping
vision.py	Multi-engine analyzer (local OCR / Claude / OpenAI / Gemini / Ollama)
parser.py	BeautifulSoup HTML parsing
extractor.py	Orchestrates scrape → vision/OCR → merge → save

Error handling

• Page load failure → up to 3 retries
• Per-card crop failure → skip and continue
• Vision/OCR failure → fall back to HTML parser
• Selector auto-detection failure → log and exit cleanly with empty results