GTM Analytics

No. 4 of 30 ยท Open source and self-hosted

Umami

The open-source cookieless tool that is genuinely pleasant to self-host.

Original abstract illustration in the form of an area chart, drawn for this review
The one fact worth knowing

Its hosted free tier gives you 100,000 events a month with no badge and no time limit.

The verdict

Umami is the open-source cookieless tool I would actually recommend someone self-host, because unlike Matomo it does not ask you to become a systems administrator to do it. It is a Node application over Postgres or MySQL, it deploys cleanly to the usual platforms, and it is small enough to read. The script is about two kilobytes and sets no cookies.

It has also grown up commercially without spoiling the open-source version. The hosted Hobby tier is free at 100,000 events a month with one site and six months of retention, no badge required, which is the most usable free hosted tier in the cookieless category. Pro at $20 gives a million events, 20 sites and two years of retention.

What you actually get

The Business tier at $200 a month is where it gets interesting and where most readers will stop reading: 10 million events, session replays and heatmaps included, white labelling and a streaming API. That is expensive against Absolutely Analytics or Pirsch at the same volume, and the recording allowance of 5,000 replays is modest. But for a company that wants behaviour tooling inside an open-source stack it is one of very few options.

The honest limits are depth and accuracy. Funnel analysis, retention and breakdowns exist and are competent rather than rich, segmentation is thinner than Plausible's, and the bot filtering is less aggressive than the tools that make a feature of it. The data score of 10 is straightforward: it is MIT-licensed software running on your database, and nothing here is more yours than that.

Best for

Developers and small teams who want cookieless analytics they can host themselves without operational drama, or a genuinely free hosted tier that does not demand a badge.

Getting it running

Docker or a Node deploy against Postgres or MySQL, which is an afternoon rather than a project. Hosted cloud is a script tag.

What to watch out for

Depth is modest: segmentation and funnel analysis are thinner than Plausible's or Pirsch's. Bot filtering is less developed than the tools that advertise it, so treat raw counts with more caution. Session replay and heatmaps only appear on the $200 Business tier. Self-hosting still means you own backups and upgrades.

Pricing and practicalities

Self-hosted free under MIT. Cloud Hobby free at 100,000 events a month, one site, six months retention. Pro $20 at 1M events, 20 sites, two years. Business $200 at 10M events with replays and heatmaps. Additional events are metered per event. Prices read 2026-09-11.

Who should go elsewhere

Teams who want depth and polish rather than ownership, where Plausible Analytics and Pirsch Analytics are better daily tools for less than the Business tier. Also not for anyone who wants heatmaps cheaply, where Absolutely Analytics includes them from $14.

How it compares

Against Matomo it is the much lighter self-hosting story with a fraction of the features, and for most small teams that is the better trade. Against GoatCounter it is the more capable and less austere of the two open-source options.

The ten measures

Each measure is scored 0 to 10 and weighted. The overall figure is the weighted mean, and it is relative to the other 29 tools on this list rather than absolute. How this works.

  • Insight quality ×37
  • Privacy architecture ×29
  • Value for money ×29
  • Setup and time to data ×1.56
  • Script weight ×1.59
  • Analytical depth ×16
  • Behaviour tooling ×14
  • Data ownership ×110
  • Accuracy ×16
  • Vendor conduct ×17

What this rests on

Sources

Umami's own pricing page and feature matrix, read 2026-09-11, including tier allowances, retention, replay and heatmap availability and per-event overage rates; its public repository and self-hosting documentation.

Evidence

Desk research against published pricing and documentation plus the public source repository. I have not run a self-hosted instance at volume, so the setup score reflects the documented deployment path rather than an operational trial.

This is a desk review. I have not been given access by the vendor, I have no commercial relationship with them, and everything above is my opinion rather than a statement of fact. If something here is wrong, tell me and I will correct it.