Every "50 apps to self-host" list ignores the only question that matters on a 4 GB box: what does it cost to keep running. Here are the floors I work from, taken from the template catalogue rather than from marketing pages.
Comfortable on a small server (128–256 MB each)
Uptime Kuma for monitoring, a TeamSpeak server, Vaultwarden for passwords, ntfy for push notifications, FreshRSS, Mailpit for catching test email, AdGuard Home. You can run several of these together and barely notice.
Mid-weight (512 MB – 1 GB)
WordPress with its own database, Ghost, Gitea, Umami analytics, Uptime dashboards with history, n8n. n8n in particular wants 1 GB and will use it — workflow engines hold a lot in memory between steps.
Do not put these on a 4 GB box with anything else (2 GB+)
Immich, because face and object recognition runs a model. Ollama, for the obvious reason. Elasticsearch, which will take whatever heap you give it and ask for more. Anything doing video transcoding, where the constraint is CPU and not memory at all.
The arithmetic people get wrong
Add up the floors and stop at 60–70% of physical RAM, not 100%. The remainder is not waste — it is page cache for your databases, and headroom for the container that spikes during a backup. A box running at 95% memory does not fail gracefully; it fails at 3am when a nightly job allocates.
Second mistake: forgetting the database that comes with the app. WordPress is not 512 MB, it is 512 for PHP plus whatever MariaDB takes. Multi-service templates deploy both, which is convenient and also doubles the footprint you were budgeting for.
What is everyone running on their smallest box? Curious where the practical ceiling actually lands.
Comfortable on a small server (128–256 MB each)
Uptime Kuma for monitoring, a TeamSpeak server, Vaultwarden for passwords, ntfy for push notifications, FreshRSS, Mailpit for catching test email, AdGuard Home. You can run several of these together and barely notice.
Mid-weight (512 MB – 1 GB)
WordPress with its own database, Ghost, Gitea, Umami analytics, Uptime dashboards with history, n8n. n8n in particular wants 1 GB and will use it — workflow engines hold a lot in memory between steps.
Do not put these on a 4 GB box with anything else (2 GB+)
Immich, because face and object recognition runs a model. Ollama, for the obvious reason. Elasticsearch, which will take whatever heap you give it and ask for more. Anything doing video transcoding, where the constraint is CPU and not memory at all.
The arithmetic people get wrong
Add up the floors and stop at 60–70% of physical RAM, not 100%. The remainder is not waste — it is page cache for your databases, and headroom for the container that spikes during a backup. A box running at 95% memory does not fail gracefully; it fails at 3am when a nightly job allocates.
Second mistake: forgetting the database that comes with the app. WordPress is not 512 MB, it is 512 for PHP plus whatever MariaDB takes. Multi-service templates deploy both, which is convenient and also doubles the footprint you were budgeting for.
What is everyone running on their smallest box? Curious where the practical ceiling actually lands.