Demo mode. Sample data, Gmail sends are dry runs (the exact API request is recorded, nothing leaves), no login. The sample data resets once a day, or now with the button.

Third-party services and costs

The system is designed to run on things you already own. After handoff nothing routes through the developer's accounts, servers, keys or subscriptions.

Service Needed for Owner Cost
Google Cloud project with the Gmail API enabled OAuth client and API access you no charge for the Gmail API
The dedicated Gmail account sending you your existing account; a Google Workspace seat is optional (it makes the OAuth app Internal and lifts the daily limit to 2,000)
A server hosting the admin and the scheduler you any small Linux VPS (1 vCPU, 1 GB RAM) or a machine you already run; the app idles at well under 200 MB of RAM
Domain name and TLS certificate HTTPS for the admin and the OAuth callback you a domain you own; certificates are free through Let's Encrypt (Caddy does this automatically)
Anthropic API key (optional) the model fallback for page layouts no heuristic fits you pay per use, no subscription; a fallback call sends at most 14,000 characters of page text, so with Claude Opus 5 at $5 per million input tokens and $25 per million output tokens it costs roughly two to five cents per page, and it only runs on pages that need it
Python packages runtime open source free (FastAPI, SQLAlchemy, BeautifulSoup, httpx, APScheduler, openpyxl, google-auth)

Nothing else. No SaaS, no per-seat licence, no metered scraping service, no email-sending platform: the emails go straight from your Gmail account through Google's API.

Ongoing effort

A few minutes a day in the review queue. Monthly: glance at the backups folder and the run history for repeated fetch errors (a university redesigns a page, so its source needs re-adding). Yearly: a restore test.