I don’t want Google Analytics on this site. I don’t want any third-party analytics script that phones home, sets cookies, or ends up in some ad network’s data pipeline. But I also want to know if anyone is reading what I write.

The solution is obvious in theory: track it yourself. Tonight I built it.

The Design

Five pieces:

1. A beacon script on every page that fires a non-blocking POST to /api/track
2. A Netlify serverless function that validates the request and writes to Supabase
3. A page_views table in Supabase with indexes on path and created_at
4. An /analytics page that queries Supabase at build time and renders charts
5. A Netlify edge function that puts basic auth in front of that page

The beacon sends path, referrer, user_agent, and timestamp. The function extracts the country from Netlify’s x-nf-country header — a free geolocation signal that Netlify injects automatically on every request. No GeoIP lookup, no extra service.

The analytics page builds once per day (the site already has a daily Netlify rebuild). It shows total views, views today, views last 7 days, a 30-day line chart, a top-10-pages bar chart, and a referrer breakdown grouped by domain. Chart.js via CDN, dark mode aware, same font and color scheme as the rest of the site.

The Bugs Worth Mentioning

The beacon was silently doing nothing. I put the tracking script in Layout.astro. It compiled fine, deployed fine, and fired on exactly zero page loads — because none of the pages actually use Layout.astro. They all use BaseHead directly. Classic case of writing to the right abstraction for the wrong reason. Moving the script to BaseHead.astro fixed it immediately.

There was a second wrinkle: this site uses Astro’s ClientRouter for view transitions, which means pages don’t fully reload on navigation. A regular <script> tag only fires on the initial load. The fix is to listen for astro:page-load instead, which fires on both initial load and every client-side navigation:

document.addEventListener('astro:page-load', () => {
  fetch('/api/track', { method: 'POST', keepalive: true, ... });
});

The edge function auth loop. Protecting /analytics with basic auth via an edge function sounds simple. It isn’t, because of trailing slashes. Astro builds the page at /analytics/index.html, so the canonical URL is /analytics/. If you put the edge function on /analytics (no slash), it passes auth, calls context.next(), Netlify 301-redirects to /analytics/, and the browser drops the Authorization header on the redirect. A second auth challenge fires, the browser retries, Chrome hits its retry limit, and you get ERR_TOO_MANY_RETRIES with no dialog ever appearing.

The fix: only guard /analytics/. Netlify’s trailing-slash normalization happens before the edge function runs, so the redirect is transparent and the auth challenge fires exactly once on the canonical URL.

Chrome cached credentials. After debugging the loop, I was still getting ERR_TOO_MANY_RETRIES even after the fix deployed. Turns out Chrome had silently cached the bad credentials from the broken earlier deploys and was auto-retrying them on every request, never surfacing a dialog. Opening an incognito window confirmed the fix was working. Cleared the cached credentials and normal browsing worked too.

The Honest Take

The whole thing — design, implementation, debugging, deploy — took one session. The interesting problems were not the ones I expected. I thought the hard part would be the Supabase integration or the Chart.js charts. Those were fine. The bugs were in the wiring: a layout component nobody uses, a script that doesn’t survive view transitions, a redirect that strips auth headers.

These are exactly the kind of subtle environmental issues that are hard to anticipate in advance and fast to fix once you see them. I saw them because the feedback loop was tight — deploy, test, observe, fix, repeat. Claude Code kept the whole context in one place so I didn’t have to.

The dashboard is live. The data is mine. No cookies, no third parties, no tracking I don’t control.

---

korbonits.com is my personal blog. I write about ML, software, and books.