I Built a Self-Hostable Plausible + Sentry Alternative in One Day - That Runs on Shared Hosting
I worked on projects that run on shared hosting. PHP + MySQL, no root access. It's a real environment that a lot of developers actually ship to - and almost no tooling is built for it.
But every analytics or error tracking tool I looked at assumed you had a VPS at minimum, or were happy paying a SaaS bill every month. Plausible is great - but it's paid per site, and self-hosting it means Docker, which means a VPS. Sentry's free tier is generous until it isn't. And there was a hard requirement: no third-party services touching user data.
So I built one. In a day. It's called Micrologs.
What it does
Drop one script tag on your site:
<script
src="https://yourdomain.com/snippet/micrologs.js"
data-public-key="your_public_key"
data-environment="production"
async>
</script>
From that point, you get:
- Pageviews, sessions, unique visitors, bounce rate
- Country / region / city breakdown (via local MaxMind GeoLite2 — no runtime API calls)
- Device, OS, browser breakdown
- Referrer categorization — organic, social, email, referral, direct
- UTM campaign tracking
- JS errors auto-caught —
window.onerrorandunhandledrejection, grouped by fingerprint - Manual error tracking from any backend over a single HTTP call
- Audit logging
- Tracked link shortener with click analytics
All of it hits your own database. Nothing leaves your server.
The constraint that shaped everything
Shared hosting means no Redis, no background workers, no daemons, no WebSockets. You get PHP and MySQL and that's it.
This forced some interesting decisions.
Rate limiting without Redis. Most rate limiters use Redis or a DB table with a timestamp. I went file-based — append-only .req files, one per request, inside a per-IP directory. Counting recent requests = counting files newer than the window. No locking, no DB writes on every request, works on any host.
// Count recent attempts — each file is one request
foreach (glob($userDir . "/*.req") as $file) {
if ($now - filemtime($file) <= $windowSeconds) {
$attempts++;
} else {
@unlink($file); // clean up expired ones while we're here
}
}
Cleanup runs probabilistically — 1% chance on each request. No cron job needed.
Geolocation without API calls. MaxMind's GeoLite2 database ships as a .mmdb file you drop on your server. Every lookup is local. Zero latency overhead, zero external dependency at runtime.
Visitor identification without being creepy. No raw IPs stored, ever. IPs are SHA-256 hashed with a salt immediately on ingestion. For visitor tracking, a UUID is stored in a cookie for 365 days. If the cookie gets cleared, a canvas fingerprint kicks in as a fallback to re-identify the same visitor — then re-associates the new cookie hash so the next visit is seamless again.
The architecture is designed in stages
v1 is a clean REST API — works on shared hosting, zero extra infrastructure.
v2 will add Redis caching, async queuing, and webhook alerts — opt-in for people who have a VPS.
v3 will add WebSockets and a live dashboard feed — opt-in for people who want realtime.
The key decision: each stage is strictly opt-in. Shared hosting users will never be broken by what VPS users unlock. v1 will always work on v1 infrastructure.
Some implementation details worth talking about
Error grouping by fingerprint. This is how Sentry works and it's the right approach. When an error comes in, I hash project_id + error_type + message + file + line into a SHA-256 fingerprint. Same error fires 1000 times — 1 group, 1000 occurrences. If you mark an error as resolved and it fires again, it automatically reopens.
$fingerprint = hash(
"sha256",
$projectId . $errorType . $message . $file . ($line ?? "")
);
Public key vs secret key separation. The JS snippet uses a public key — safe to expose in the browser, locked to a whitelist of allowed domains. Analytics queries and link management use a secret key — server-side only, never in frontend code. One install supports multiple projects, each with their own key pair.
SPA support in the snippet. History-based routing doesn't trigger page loads, so pushState and replaceState get patched to fire pageviews on navigation. popstate and hashchange are also handled.
history.pushState = function () {
_push.apply(this, arguments);
onUrlChange();
};
Bot filtering. UA string matching is obvious, but real browsers always send Accept-Language and Accept headers. If those are missing, it's a bot or a script regardless of what the UA says.
What the API looks like
curl https://yourdomain.com/api/analytics/visitors.php?range=30d \
-H "X-API-Key: your_secret_key"
{
"success": true,
"data": {
"unique_visitors": 1842,
"total_pageviews": 5631,
"total_sessions": 2109,
"bounce_rate": 43.2,
"over_time": [
{ "date": "2026-01-28", "pageviews": 178, "unique_visitors": 91 },
{ "date": "2026-01-29", "pageviews": 204, "unique_visitors": 113 }
]
}
}
It's an engine, not a dashboard. The data comes back as JSON — what you do with it is up to you. Build a dashboard, pipe it into Grafana, query it from your admin panel, whatever fits your stack.
Stack
- PHP 8.1+
- MySQL 8.0+ / MariaDB 10.4+
- MaxMind GeoLite2 (local
.mmdbfile) - Vanilla JS snippet, zero dependencies, ~3KB
No Node, no Docker, no Redis, no build step. Clone it, import the schema, fill in the env file, drop the snippet. That's the entire setup.
It's already in production
We shipped it internally at the company the same day I built it. It's tracking real traffic right now. That's the other reason for the shared hosting constraint — it wasn't a hypothetical, it was the actual target environment.
Open source, MIT
GitHub: github.com/OmDongaonkar03/Micrologs
If you're on shared hosting, running a privacy-first product, or just tired of paying for tools you could own — give it a try. Issues and PRs are open.
v2 is next. Webhooks for error alerts, Redis caching as opt-in, async queuing. If you have thoughts on what matters most, open an issue.
SOCIAL SHARE CARD GENERATOR