29 self-hosted apps idle above 1 GB, and 164 stay under 256 MB

copySelfHost Atlas (2026). 29 self-hosted apps idle above 1 GB, and 164 stay under 256 MB. https://selfhostatlas.com/blog/heaviest-and-lightest-self-hosted-apps. Data: doi:10.5281/zenodo.22983606

We measured idle RAM for 268 self-hosted apps, one app at a time on a server pruned of earlier apps. The spread is wide: the lightest apps idled at 3 MB and the heaviest at 16,824 MB.

#The shape of the distribution

• The 164 light apps have a median idle of 94 MB. Nineteen idle at 10 MB or less.
• The 29 heavy apps are 11% of the apps measured and use 59% of all the memory in the set (about 79,800 of 135,309 MB).
• The five heaviest alone account for 30% of it.
• All 164 light apps added together use 15,640 MB, which is less than PostHog on its own (16,824 MB).

#The 29 above 1 GB

The other nineteen range from 2,095 MB (Tube Archivist) down to 1,034 MB (Invoice Ninja). In our category taxonomy, the only category with more than two apps in the heavy group is no-code databases (Teable, NocoDB, Baserow).

#The lightest ten

#What these numbers are

Idle RAM is the memory docker stats reports for every running container of the app, summed, 30 seconds after the install finishes. That is a snapshot, not steady-state usage. It includes the app's own database and helper containers, because those are part of what you run. It is not a peak, not under load, and not a sizing recommendation.

262 of the 268 measurements were taken on a Google Cloud e2-standard-2 with Ubuntu 26.04 LTS and Docker, on a machine pruned of earlier apps. Six were run on larger machines: PostHog (e2-standard-8), and Sentry, Langfuse, RustDesk, AppFlowy and LiteLLM (e2-standard-4). We did not test whether the host size changes an app's idle figure, so read each entry as measured on that machine.

How the harness works is described on How we test. If you are planning a server, the best VPS by workload pages start from the minimum RAM we list, and the listed-versus-idle post explains why that number and this one differ.

#About the numbers

This post uses the install-verification dataset exported on 2026-10-25 (release v1.8.0: 312 records, 268 with a measured RAM figure, measured between 2026-08-28 and 2026-10-25). The dataset is CC BY 4.0 and cited as concept DOI 10.5281/zenodo.22983606, which always resolves to the latest version.