Why self-host analytics at all

The privacy angle is the obvious one: first-party data that never leaves your own infrastructure, no GDPR consent-banner requirement in most interpretations (since there’s no third-party cookie or cross-site tracking involved), and nothing to disclose to a company whose core business model is building ad profiles. The practical angle matters just as much, though. A privacy-respecting analytics script is small, often under a few KB, and isn’t on the ad-blocker and tracker-blocker lists that silently zero out a meaningful slice of Google Analytics’ numbers. What you see in a self-hosted dashboard is closer to the truth of your actual traffic, not traffic minus everyone running uBlock Origin or Pi-hole.

Plausible: lightweight, opinionated, and genuinely simple

Plausible is built around one idea: give you the handful of numbers that actually matter (visitors, page views, referrers, top pages, bounce rate) without a sprawling feature set to configure. The tracking script is tiny, around 1KB, cookieless by design, and doesn’t fingerprint visitors, it counts unique visitors using a rotating daily salt hash of IP + user agent instead of a persistent identifier. That design choice is also why Plausible can credibly claim GDPR/CCPA compliance without a cookie banner in most legal readings, though “most readings” isn’t the same as “guaranteed everywhere,” so don’t take that as formal legal advice for your specific jurisdiction.

Self-hosting it is a Docker Compose stack: the Plausible app itself, a ClickHouse database for the actual event storage, and Postgres for user/site config. ClickHouse is the heavier piece here, it’s a real analytical database built for fast aggregation over large event volumes, and while it’s not as bulky as a Matomo install, it’s more than Umami’s footprint. Budget something in the 1-2GB RAM range comfortably for a low-to-moderate traffic personal site or small blog network.

The tradeoff is depth. Plausible deliberately doesn’t do session recordings, heatmaps, funnel analysis, or granular custom event dashboards out of the box (custom events exist, but they’re basic counters, not a flexible event-property system). If what you want is “how many people visited, where did they come from, what did they read,” Plausible answers that cleanly and gets out of your way. If you want to analyze behavior in depth, it’s the wrong tool.

Umami: the lightest option, built for a single operator

Umami takes the same cookieless, privacy-first philosophy as Plausible but trims the backend further. It’s a single Node.js application backed by Postgres (or MySQL), no ClickHouse, no separate analytical database layer, which makes it noticeably lighter to run. A small VM or LXC with 512MB-1GB RAM handles a personal site’s worth of traffic without strain, and the whole stack is one app container plus one database container.

Functionally it covers the same core ground as Plausible: visitors, page views, referrers, device/browser/OS breakdowns, and a basic custom-events system that’s actually a bit more flexible than Plausible’s (you can attach simple key-value data to events). The dashboard is clean and the multi-site support is solid if you’re tracking more than one domain from one instance, genuinely useful if you’re running rhomelab.com, a Lemmy instance, and a Listmonk signup page all under one roof.

The honest tradeoff against Plausible is maturity and ecosystem. Plausible has been around longer, has a larger community, and a hosted/paid option if you ever want someone else to run it for you. Umami is community-driven open source without an official paid-hosting arm (third parties offer hosting, but it’s not the core project’s business model), which is fine for a homelab operator who’s self-hosting by default anyway, but worth knowing if you ever want a vendor to call when something breaks.

Matomo: the Google Analytics replacement that acts like Google Analytics

Matomo (formerly Piwik) is a different category of tool entirely. It’s built to be a near feature-complete alternative to Google Analytics or even Adobe Analytics, not a minimalist privacy-first dashboard. That means session recordings, heatmaps, funnel analysis, A/B testing, e-commerce tracking, a full goals/conversions system, and a granular permissions model if multiple people need different levels of access to the data. If you’re running the Proxmox Starter Kit storefront or anything with a real checkout funnel you want to analyze step by step, Matomo is the only one of the three built for that out of the box.

That completeness comes at a real resource and complexity cost. Matomo runs on a PHP/MySQL (or MariaDB) stack, and a real install with any meaningful traffic wants 2GB+ RAM and a database that will grow substantially larger than Plausible’s or Umami’s for the same traffic volume, because Matomo stores far more granular data per visit by default (archiving/aggregation jobs that summarize old data need to run via cron or they’ll eventually slow the UI down). It also defaults to first-party cookies for session tracking, which you can disable for cookieless/anonymous mode, but that’s a deliberate configuration step, not the out-of-the-box behavior the way it is for Plausible and Umami.

Matomo also has a feature neither of the others does: an opt-in tracking API compatible enough with Google Analytics’ analytics.js/gtag.js patterns that some migration tooling exists, and a WordPress plugin if that’s part of your stack. If part of the draw is replacing an existing Google Analytics setup with minimal disruption to how you think about the data, Matomo is built with that migration in mind in a way the other two aren’t.

Comparing the practical dimensions

Resource footprint: Umami is the lightest (one app + one database, works fine on 512MB-1GB). Plausible is moderate (ClickHouse adds real but manageable overhead, plan for 1-2GB). Matomo is the heaviest (PHP/MySQL stack plus cron-driven archiving, plan for 2GB+ and a database that keeps growing).

Feature depth: Matomo is a full analytics suite, heatmaps, session recordings, funnels, goals, e-commerce. Plausible and Umami both intentionally stop at traffic/referrer/basic-event numbers, the 80% most people actually look at.

Cookie/privacy posture out of the box: Plausible and Umami are both cookieless by default. Matomo defaults to first-party cookies and needs an explicit configuration change to run cookieless, which also affects whether you need a consent banner depending on your jurisdiction and configuration.

Setup complexity: Umami is the simplest Docker Compose stack of the three. Plausible is a few more containers (ClickHouse adds a step) but still a known, documented Compose file. Matomo is the most involved, PHP-FPM, a web server in front of it, MySQL, and a cron job for archiving that people forget to set up and then wonder why the dashboard gets slow.

Multi-site support: All three handle multiple sites from one instance. Umami’s multi-site switcher is particularly clean for a small handful of properties under one operator.

Picking one for your homelab

If you want the simplest possible setup and just need basic traffic numbers: Umami. Lightest footprint, fastest to stand up, does the core job well.

If you want a slightly more established project with the same minimalist philosophy and don’t mind the extra ClickHouse container: Plausible. Same cookieless privacy posture as Umami, bigger community, a hosted option exists if you ever want to offload running it.

If you need real funnel/conversion analysis, heatmaps, or you’re specifically replacing an existing Google Analytics setup and want feature parity: Matomo, on hardware that can carry it. Don’t run it on the same tiny LXC where you crammed Umami, it wants real resources and an actual maintenance routine (the archiving cron job isn’t optional in practice).

Bottom line

All three get you out of Google’s ad-data pipeline and back to first-party, self-hosted numbers you actually control. The decision isn’t really “which one is best,” it’s “how much analytics depth do I actually use.” Most personal sites and small homelab projects only ever look at visitor counts, top pages, and referrers, in which case Umami or Plausible answers the real question with a fraction of Matomo’s resource and maintenance cost. Reach for Matomo only when you have an actual funnel or conversion flow worth analyzing in that much depth, not because it has the most checkboxes on a feature comparison page.