Added
- Every job is shared across a 4-machine scraping cluster, with fair turns between users so a small job never waits behind a large one
- New CSV columns: star rating (e.g. 4.5) and review count
- Proxy data meter on the dashboard, with a warning before the monthly plan runs out
- Change your password in Settings; admins can reset anyone's password
- Admin Cluster panel showing each machine, what it's working on, and the queue
- "Partial" count on the dashboard for searches that returned data but couldn't reach every page
Fixed
- About 30% of rows were duplicates of paid listings; results now contain each listing once
- Star ratings were always blank; years in business now shows just the number; categories are separated with "; "
- "Boise, ID" was searched as two locations ("Boise" and "ID")
- Keywords containing & or # (e.g. "B&B") searched for the wrong thing
- A search blocked partway through now picks up from the page it stopped on instead of quietly returning fewer results
- Jobs could hang forever if a single search kept failing, or stay stuck on "stopping"
- Results from a timed-out batch could be silently lost
Changed
- Locations and keywords can be separated by commas or new lines
- You can delete only your own result files ("Clear" no longer removes other people's results); admins can manage all
- Only admins can replace the proxy list
- Logins lock for 15 minutes after 10 wrong passwords
- Changing your password signs you out on other devices
Added
- Kubernetes-ready containerized architecture (Docker + k3s)
- Shared-browser engine: 1 Chromium + N contexts (25x less RAM)
- Redis-backed job queue, state, auth, and log streaming
- Worker pod with automatic crash recovery
- CLI tool (
yp-cli.py) for terminal and AI agent usage
- Grok (xAI) integration replacing OpenAI for AI suggestions
- Error count tracking with live display during scrapes
- Job persistence across page reloads (localStorage)
- Documentation page and changelog
- Health check endpoint (
/healthz)
SKILL.md for Claude Code AI agent integration
Performance
- 249 concurrent scrapes in 3GB RAM (was 45GB with old architecture)
- 3,100+ businesses/min with proxy rotation
- Web container: 83MB idle, Worker: 120MB idle
- 250 searches complete in ~30-45 seconds
Fixed
- Open redirect vulnerability on login
- Path traversal in file downloads
- SSE log stream infinite loop with no timeout
- Jobs stuck in "running" after worker crash
- Browser context memory leaks on exceptions
- Redis connection loss causing permanent worker death
- Double-click race condition on Start button
- Page reload losing track of running job
Changed
- Replaced SQLite with Redis for user/settings storage
- Replaced file-based state with Redis pub/sub
- Admin password moved from hardcoded to environment variable
- Input validation: max 200 keywords/locations, bounded numerics
- File upload limited to 10MB
Added
- Public landing page with feature overview
- Hyperspace starfield animation during scrapes
Added
- Parallel scraping with configurable worker count
- Admin panel for user management
- User authentication with Flask-Login
- Settings page for API key configuration
- AI-powered keyword/location suggestions (OpenAI)
- Proxy rotation with health tracking
- Chunked CSV output for large result sets
- Business limit to cap results
- Auto-cleanup of old result files
Added
- Initial release
- Web UI for configuring and running scrapes
- Playwright-based scraping engine
- Real-time log monitoring via SSE
- CSV export with download
- Proxy file support
- Gunicorn production server with systemd service