Self-Hosted vs Hosted Privacy Analytics in 2026

Open-source analytics is free as a license and a job as a service. Here is the ledger for self-hosting Plausible Community Edition, Umami or Matomo against hosted tools, and a ten-minute way to decide.

By Null Agency · Updated November 4, 2026 · Written for developers and small teams choosing where their analytics should run

TL;DR: self-hosting buys custody, and bills you in maintenance

Self-hosted analytics trades a subscription for a recurring job. Plausible Community Edition, Umami and Matomo are well-regarded open-source tools, and running them yourself gives you the strongest possible answer to "where does the data live?". It also makes you the on-call engineer for a system whose failure is silent.

Related reading: our cookieless analytics comparison covers the feature side; this page is about where the thing runs.

What "self-hosted" actually means

Every tool in this category has the same three moving parts: a tracking script that runs in your visitors' browsers, a collector that receives the events, and a database plus dashboard that stores and displays them. A hosted service runs all three for you. Self-hosting means the collector, the database and the dashboard are yours, and only the script is a snippet on your pages.

The concrete stacks differ. Umami is a Node.js application backed by PostgreSQL or MySQL. Matomo is a PHP application backed by MySQL or MariaDB, and it is the heaviest of the group because it is the broadest: it is a full analytics suite in the Google Analytics mould, with the surface area that implies. Plausible's self-hostable Community Edition is the open-source release of Plausible; its stack is documented by the project, and what the Community Edition includes compared with the paid hosted product is something to read in the current docs before you commit.

None of that is exotic for an engineer. The point is that each piece is a thing that ages: operating-system patches, runtime versions, database versions, certificates, disk usage. A tracking pixel on a hosted service has none of those on your plate.

The ops ledger: what you take on

This is the table most "best analytics" lists skip. It does not claim any tool needs a particular amount of time; it lists the jobs. Count how many of them you already do for another service.

JobSelf-hostedHosted service
ProvisioningYou size and pay for a server or container, plus a databaseCreate an account, paste a script tag
TLS and domainYou terminate HTTPS and renew certificates for the collector hostnameVendor handles it
UpgradesYou read release notes, run migrations, roll back if one failsVendor ships them
BackupsYou schedule, store and test restores of the databaseVendor's responsibility; you can still export your own rows
Uptime monitoringYou watch the collector endpoint and alert yourselfVendor monitors; you check their status page
ScalingYou resize when traffic grows or the database slows downPriced into the plan
Security patchingYou track advisories for the app, runtime, database and OSVendor's job
Data locationWherever your machine isWherever the vendor runs it
Source accessFull, for open-source toolsDepends on the vendor

Two rows deserve emphasis. The first is backups: an untested backup is a hope. If your analytics history matters, the restore drill is part of the cost. The second is uptime monitoring, because analytics fails differently from a website.

Why analytics outages are silent

When your site goes down, someone emails you within the hour. When your analytics collector goes down, nothing visible breaks. The script on each page fires a request once per pageview; if the collector does not answer, that pageview is gone, with no retry queue on your side and no error shown to the visitor.

The result is a dashboard with a flat stretch you only notice when you compare a launch week to the week before. For a hobby site that is a shrug. For a campaign report that is a missing number and an awkward conversation. The mitigation is cheap and easy to forget: an external uptime check on the collection endpoint (not just on the dashboard login page), and an alert that goes somewhere you will see it.

Hosted vendors have outages too. The difference is that someone else is paged, and a status page tells you the window to annotate. With self-hosting, you are the status page.

Data location and custody

This is the strongest argument for self-hosting, and it is a real one. If a policy or a regulation says personal or business data must stay on infrastructure you control, or in a specific country, or that no third-party processor may touch it, then a self-hosted tool in your own environment answers the question directly. No hosted vendor, including us, can match that by definition.

Two cautions keep the argument honest. First, custody is not the same as privacy-by-design. What a tool collects (IP addresses, identifiers, cookies) is a product decision; where it stores them is a hosting decision. A cookie-based tool on your own server is still a cookie-based tool. Our explainer on cookieless tracking covers what a design that avoids identifiers looks like, whoever hosts it.

Second, hosted does not mean careless. Plausible, for example, publishes that its hosted service processes data in the EU only, and Fathom offers an EU-isolation option. GhostMetrics runs on Cloudflare's edge and stores no cookies, no localStorage and no raw IP addresses. If your requirement is "no personal data stored" rather than "data on my own disk", a hosted cookieless tool can satisfy it, and you should state your requirement that way when you evaluate.

The cost model, qualitatively

Prices change, so this section avoids numbers for third parties and describes the shape of each model. For current figures, read each vendor's pricing page.

OptionSource modelWhere it runsHow you pay
Umami (self-hosted)Open source (MIT)Your server and databaseInfrastructure and your time; no license fee
Matomo (self-hosted)Open sourceYour PHP and MySQL environmentInfrastructure and your time; a hosted Matomo Cloud also exists
Plausible Community EditionOpen source, self-hostableYour serverInfrastructure and your time; hosted Plausible is a separate paid product
Plausible (hosted)Open source code, hosted by the vendorVendor, EU-only processingTrial then paid plans scaled by monthly pageviews
FathomProprietaryVendor; EU-isolation optionPaid, pageview-tiered; no free plan
GhostMetricsProprietary, not self-hostableCloudflare edge (Workers and D1)Free for one site; Pro $9.99/mo flat with unlimited pageviews

The practical difference is what scales your bill. Self-hosting scales with your infrastructure and your attention: a quiet site fits on a small machine, a busy one needs a bigger database and more care. Pageview-tiered hosted plans scale with traffic. A flat price scales with neither. Neither shape is better in the abstract; it depends on whether your traffic is growing, spiky or steady, and whether your time is the scarce resource.

Upgrades, backups and the long tail

The first install is the easy day. The cost lives in year two.

Upgrades

Open-source projects ship fixes and features continuously, and a self-hosted instance that falls several versions behind becomes progressively harder to bring current, because migrations stack. Teams that do this well put the upgrade on a calendar. Teams that do it badly discover the version gap during an incident.

Backups

A nightly database dump copied off the machine is the minimum. Test one restore per quarter to a scratch machine; if you cannot rebuild the dashboard from the dump, you do not have a backup. Hosted vendors do this work for you, though it is still reasonable to keep your own copy. GhostMetrics offers a raw CSV and JSON export of every stored row on every plan for that reason.

Storage growth

Event tables grow with traffic. Self-hosted instances need a plan for disk usage and for query speed on large date ranges, and some projects document retention or aggregation options that you would configure yourself.

Security

An analytics dashboard is a login page on the public internet in front of a database. Keep the app, the runtime and the operating system patched, put the dashboard behind strong authentication, and limit who can reach the database directly. This is ordinary hygiene; the point is that it is yours.

Who should self-host, and who should not

Self-host if a data-residency rule, a procurement policy or a contract forbids a third-party processor; if you need to read or modify the source; if you already operate similar services and the marginal cost of one more is small; or if running your own infrastructure is part of what you want to do. For these teams the extra work is the product.

Go hosted if you want answers to "which pages work and where does traffic come from" and nothing else; if nobody on the team wants to be paged for a chart; or if you run many small sites and would be installing the same stack repeatedly. For a single personal or small business site, the free tier of a hosted cookieless tool costs less than the first hour of setup.

Hybrid is legitimate. Some teams self-host for the regulated product and use a hosted tool for marketing pages. Nothing requires one choice across everything you run, and our static site analytics guide shows how little a script-tag install asks of a site.

Where GhostMetrics fits

GhostMetrics is strictly the hosted option. We say so in the table above because the comparison is only useful if it states where we do not apply: not open source, not self-hostable, no EU-only processing guarantee. What it offers instead is a script of about 3 KB compressed, a free plan with full history, and no machine to look after.

Decide in ten minutes

Answer these in order. Stop at the first one that gives you a hard answer.

  1. Write down your hard constraint, if any. "Data must stay on our infrastructure" or "must be open source" or "must be in a named country." If you have one, shortlist only tools that meet it, and self-hosting is probably on the list.
  2. List the services you already patch and back up. If you have a working routine for two or more, adding one more is cheap. If the list is empty, the first one costs you the whole routine.
  3. Estimate your traffic shape. Steady and small: nearly any option works. Growing fast: hosted plans scale without your involvement; self-hosting scales with your attention.
  4. Decide who gets paged. If the answer to "who notices when collection stops?" is nobody, choose hosted or set up an external check on the collector before you start.
  5. Check what you need beyond pageviews. Custom events, UTM reports, public dashboards, Core Web Vitals and export differ by tool and by edition. Confirm each on the vendor's current docs, and for self-hosted editions confirm they are in the edition you will run.
  6. Try the hosted option first if you are unsure. A free plan is a reversible experiment: install the tag, look at a week of data, and export it if you leave. A server you built is harder to walk away from.
  7. Write down the exit. Whatever you choose, note how to export your rows and what the second tool would be. History does not transfer between tools automatically, so the exit plan is the export plan.

If steps 1 to 3 leave you without a hard constraint and without a spare operator, the answer is hosted. If they leave you with a constraint, the answer is the open-source tool that meets it, plus the monitoring described above.

Our recommendation

Pick by the failure you can tolerate. A self-hosted instance fails by consuming your time and, if neglected, by going dark. A hosted tool fails by costing money each month and by holding data on someone else's infrastructure. Both are acceptable; the mistake is choosing one without pricing the other.

For most small teams, the sensible default is a hosted cookieless tool with a free tier and a clean export, revisited if a custody requirement appears. For teams that already have the requirement, self-host an open-source tool and add the collector monitoring first, not last. The full list of related guides is in our analytics guides.

Hosted, cookieless, and nothing to patch? Try GhostMetrics

Hosted, zero ops$0 forever1 site free$9.99/mo flat Pro

GhostMetrics is the hosted end of this comparison: one script tag, no server, no database, no upgrades, running on Cloudflare's edge. It is not open source and not self-hostable, so if a license or a residency rule forces you to own the stack, pick one of the self-hosted tools above. If you just want the numbers, the free plan is free forever: one site, unlimited pageviews, real-time dashboard, every view, full history.

Need several sites or a public dashboard? Pro is $9.99/mo flat, unlimited pageviews, unlimited websites (fair-use ceiling of 250), with a 30-day free trial (card required to start, nothing charged for 30 days, cancel anytime). Raw CSV or JSON export is on every plan, including free, so leaving is always possible.

Start free — no server, no card See the live demo

FAQ

Is self-hosted analytics really free?
The software license is free for Plausible Community Edition, Umami and Matomo. You still pay for the server, the database, backups and your own maintenance time. Check each project's current docs for which features the self-hosted edition includes.
Does self-hosting make analytics more private?
It changes who holds the data, not what is collected. A self-hosted tool keeps every row on a machine you control, which helps with strict data-residency or no-third-party policies. A hosted cookieless tool can be just as light on personal data; the difference is custody.
What happens to my analytics if my server goes down?
Pageviews sent while the collector is unreachable are usually lost, because the browser fires the request once and moves on. A self-hosted setup therefore needs monitoring on the collection endpoint, not just the dashboard.
Can I move from self-hosted to hosted later?
Yes, with a caveat: history does not transfer automatically between tools. Export your old data first, run both tools in parallel for a couple of weeks, then retire the old one. GhostMetrics gives raw CSV or JSON export on every plan for the return trip.
Who should self-host analytics?
Teams that must keep data on their own infrastructure, teams that already run and patch similar services, and people who enjoy it. Everyone else is usually better served by a hosted tool whose pricing they can read in advance.

Umami Alternatives

What self-hosting Umami costs, and hosted picks

Matomo Alternatives

GA-style depth without the self-host overhead

Plausible Alternatives

Open-source and hosted options compared

GhostMetrics Live Demo

The real dashboard, live data

Disclosure: GhostMetrics is built by Null Agency and we use it on this site, so we're not a neutral party about our own product — we've kept the claims to what the free and paid plans actually do, and pointed to Umami and other open-source, self-hostable tools where they genuinely fit better. Pricing for third-party products can change; always check the vendor's current pricing page before deciding.