Plausible or Umami? Self-hosted Google Analytics alternatives compared

Google Analytics is free, but it runs at Google, a US company, and sets cookies in your visitors' browsers. Plausible and Umami count visits on a server of your own, without cookies, so the visitor data stays with you. We ran both on the same small website with the same visits. What Google Analytics offers, why EU companies look for alternatives, and what you give up by leaving it, is near the end of this page, from Google's and the authorities' own pages.

The short answer

Pick Umami if you want the lighter of the two: it needed a third of Plausible's memory, showed visitors' cities, counted custom events without any setup, has a Swedish interface and is MIT-licensed. Pick Plausible if you want conversion rates for goals, scroll depth and email reports, and are fine with running ClickHouse and updating by hand twice a year.

What we tested

We installed both on the same fresh Melonslab VC-P Alloy (2 vCPU, 8 GB) with Debian 13, one after the other, by following our guides Plausible with rootless Podman and Umami with rootless Podman step by step. Each ran under a user of its own, behind the same Caddy.

Then we put one small static website behind that Caddy, with four pages and a Sign up button, and both tracking snippets in the <head> of every page. The button sent a custom event called Signup to both. In a browser, three visitors with different browsers and systems:

• came from a link on another site, read four pages and pressed Sign up;
• read two pages and left;
• read three pages and pressed Sign up.

That is 3 visitors, 9 pageviews, one visit from another site and 2 sign-ups. We compared what each dashboard showed, measured the memory after a reboot, and ran each guide's backup.

The numbers

Memory is the sum of each app's containers, measured with podman stats a few minutes after a reboot, with each dashboard opened once. Most of the data on disk is the empty databases themselves, not the test visits.

What they showed

Both counted the test the same way: 3 unique visitors, 3 visits, 9 pageviews, a bounce rate of 0 %, the same four pages and one visitor from the other site. They disagreed only on the visit duration, 7 seconds in Plausible and 4 in Umami, as each measures it in its own way.

The differences were in the details:

• Custom events. Umami listed the 2 Signup events under Events as soon as they were sent. Plausible stored them too, but showed them only after we added a goal for Signup under the site's Goals; it then counted both, with a conversion rate of 66.7 %. Plausible's setup had already created goals for outbound links, file downloads, form submissions and 404 pages, which its script tracks by itself.
• Location. Plausible showed the country, Sweden. Umami showed the country, the region and the city. Plausible needs a MaxMind account for regions and cities, as its guide explains.
• Devices. Umami recorded each visitor's exact screen size and language. Plausible records a size class such as Desktop, and no language.
• Time on page. Plausible's script also sends an engagement event when you leave a page, which gives it time on page and scroll depth. Umami counts the time between pageviews.

Within minutes of the site getting its certificate, three visitors we did not send loaded the home page, most likely crawlers that follow new certificates. Plausible counted all three, Umami two. Both drop visits from known bots and from browsers that say they are automated, so test with an ordinary browser, as both guides say.

Setup

Both install the same way: a user of their own, secrets in podman secret, a pod of Quadlet units behind Caddy. The first start of each needed no help.

Each has one step that you must not skip. Plausible lets whoever opens it first create the owner account, so its guide lets only your own address through Caddy until your account exists. Umami starts with an account admin with the password umami, so its guide has you change it over an SSH tunnel before Caddy makes it public. Both worked as written.

Plausible also needs four ClickHouse settings files downloaded from its own repository, and has no setting for trusted proxies, so its Caddy block removes four headers a visitor could otherwise use to choose their own address. Umami needs one setting, CLIENT_IP_HEADER, for the same thing.

Plausible stores the site without www.: we entered www.example.com, and it became example.com. The visits still counted, as the snippet carries the site's own ID.

Day-to-day use

Plausible shows everything for a site on one page: the numbers, sources, pages, locations, devices and goals, with filters on any of them. It is quick to read, and there is little else to find.

Umami has more menus: besides the overview, it has pages for events, sessions, real-time visitors, funnels, journeys, retention, goals, UTM tags and comparisons. It can also record session replays and heatmaps, with a separate recorder script that is off until you add it; we did not test them, and a replay records what visitors do on the page, which is more than these guides set up to collect.

Umami's interface is available in Swedish, under Settings, Preferences, Language. Plausible's is in English only.

Resources

ClickHouse makes the difference. It is a database built for billions of events, and Plausible keeps the visits in it and the accounts in PostgreSQL. Umami keeps everything in PostgreSQL. In our test, that made Plausible use about three times the memory, 650 MB against 210 MB. Plausible recommends at least 2 GB of memory for the server, and ClickHouse needs a processor with SSE 4.2.

Umami's image is the bigger one, 1.1 GB on its own, so it used more disk than Plausible's three images together.

Plausible serves its script uncompressed and lets browsers keep it for 60 seconds. Umami sends it compressed and lets browsers keep it for a day. Both scripts are small next to a typical web page.

Updates and backups

Plausible stays on the exact version in its unit file, and you update it by hand after reading the release notes, as Plausible recommends. Plausible publishes Community Edition about twice a year: 3.2.0 in January 2026 and 3.2.1 in May. Security fixes reach its hosted service first.

Umami's guide follows the 3 tag, and podman-auto-update.timer installs every new version of Umami 3 by itself. Umami released 3.3.0, 3.3.1 and 3.4.0 between August and September 2026. You get fixes quickly, without reading notes first.

Plausible's backup takes a script and a timer, because the statistics in ClickHouse and the accounts in PostgreSQL are backed up separately; our test backup was 240 KB. Umami's is one pg_dump and a key, 64 KB here. Both came back by themselves after a reboot.

Email

Plausible sends invitations, password resets and weekly or monthly reports, and needs a mail server for them, as its guide explains. Self-hosted Umami sends no email at all; Umami Cloud has email reports.

Licence and cost

We checked the projects' own pages on 14 October 2026.

• Plausible CE is open source under the AGPL 3.0, and its tracking script under the MIT licence. It is made by Plausible Insights OÜ in Estonia. According to Plausible, Community Edition does not have marketing funnels, user journeys, ecommerce revenue goals, single sign-on or the sites API, which its hosted service has, and it has only basic bot filtering. Self-hosting is free, with community support. Plausible's hosted service starts at 9 US dollars a month for 10,000 pageviews.
• Umami is open source under the MIT licence, and made by Umami Software, Inc. in the United States. Self-hosting is free, and Umami says the self-hosted version lacks email reports and the streaming API of its hosted service. Umami Cloud is free for one site and 100,000 events a month, and costs 20 US dollars a month for more.

Hosted on your own server, both keep the visitor data in Sweden, and the visitors' browsers only contact your server. Each guide shows what the app itself connects to, and how to switch it off.

Google Analytics today

What follows is from Google's own help pages, checked 2026-10-14, not from our tests: we did not set up Google Analytics for this comparison.

Sources: Google Marketing Platform, Analytics 360 compared with standard, data retention and cookie usage, all checked 2026-10-14. The retention setting does not shorten the standard reports, only explorations and funnels.

Google also says that Analytics "does not log or store individual IP addresses" from users in the EU, and that it collects their data through servers in the EU before it is forwarded to Analytics' servers for processing (Google, checked 2026-10-14).

Why EU companies look for alternatives

A summary of what the authorities have published, checked 2026-10-14. It is not legal advice: ask your data protection officer or a lawyer about your own site.

• 3 July 2023: Sweden's data protection authority, IMY, published its decisions on Tele2, CDON, Coop and Dagens industri, who sent personal data to the US through Google Analytics. None of their extra technical safeguards were enough, according to IMY. Tele2 was fined 12 million kronor and CDON 300,000 kronor, and the other three were ordered to stop using the tool; Tele2 had already stopped. IMY wrote that the decisions can guide other organisations that use Google Analytics too (IMY).
• 10 July 2023: the European Commission adopted its adequacy decision for the EU-US Data Privacy Framework. Since then, "personal data can flow freely from the EU to companies in the United States that participate in the Data Privacy Framework" (European Commission).
• 3 September 2025: the EU General Court dismissed an action to annul that decision. Its press release recalls that the Court of Justice declared the two earlier decisions for the US invalid, in Schrems I and Schrems II (Court of Justice of the EU, press release 106/25).
• October 2025: the Administrative Court of Appeal upheld Tele2's fine. IMY noted that its review covered August 2020 to May 2023, and that there is now an adequacy decision for the US (IMY).

So Google Analytics is not banned, and transfers to US companies in the framework are allowed today. But the framework is the third attempt, after two were struck down, and some companies would rather not depend on it. With Plausible or Umami on your own server in Sweden, the visitor data does not leave your server, and in our test neither stored anything in the visitors' browsers.

What you give up compared with Google Analytics

Your own analytics costs a server instead of nothing, and the data is yours. Be honest about what you leave behind:

• Google Ads. Linked to Google Ads, Analytics turns your key events into Ads conversions and builds audiences for remarketing (Google, checked 2026-10-14). Plausible and Umami count goals and events, but feed nothing back into Google's ad bidding.
• Depth. Explorations, audiences and the BigQuery export are built for large sites and analysts. Both apps here show a clear dashboard, and Plausible Community Edition lacks funnels and user journeys, which its hosted service has.
• Uptime. Google runs Analytics; when your server is down, the visits in that time are not counted, and neither app collects them afterwards. Someone of yours has to notice.
• Running it. Updates and backups are yours: Plausible by hand about twice a year, Umami by itself. Our guides make that a few commands.
• Compliance reports. No third party has audited your server. If a customer asks, you answer with your own routines.
• Your history. Your old Google Analytics data stays in Google Analytics; we did not test moving it to either app.

Which one to pick

Umami suits most small business sites: it runs in about 200 MB, shows cities without extra accounts, counts any event you send, updates itself and has a Swedish interface. Its weak points are the well-known default password, which you must change first, telemetry that is on until you switch it off, and a large image. Follow the Umami guide; it runs comfortably on a VC-P Alloy, next to other apps.

Plausible suits you if you work with goals and conversion rates, want scroll depth and time on page, or want weekly reports by email. Its weak points are the memory ClickHouse takes, countries only without a MaxMind account, the first-come owner account, and updates you do by hand. Follow the Plausible guide, on a VC-P Alloy.

Both together used under 900 MB, so a VC-P Alloy can also run both, as on our test server.