A lightweight self-hosted PaaS, with the number and its limits

The measured footprint, with its limits

On 2026-10-19, a live Levelrail control plane on a 2 vCPU, 8 GB Linux test box idled at 200 MB of resident memory (peak 253 MB) and 1.5 percent CPU, with all features at their defaults and no apps under load. That is one instance on one box, measured once. It is not a promise for your hardware, and it describes the control plane, not the apps you run on top of it.

A Mac development build on an Apple M4 Max idled at 68 MB with no apps. That is a different machine and a different build, so the two numbers do not compare and the Linux figure is the one to plan with. The method and the raw rows are in the benchmarks, and every row has a date and hardware label. Anything without one is not measured.

Why it stays small: the control plane and the node agent are each a single Go binary, state lives in SQLite in WAL mode, and ingress, TLS and HTTP load balancing use an embedded Caddy. Metrics and logs stay on the node that produced them and are queried on demand, so there is no separate monitoring stack to run.

Who it fits

• You have one or a few Linux servers, such as a VPS, a home server or a cloud VM, and want deploys from a git repository or an image tag with TLS, logs, metrics and rollback.
• You want Docker as the only container runtime and no Kubernetes to operate.
• You want the platform to leave most of a small machine to your own apps, and you are happy to check the footprint on your own hardware before you rely on it.
• You want an Apache 2.0 project with one edition, no license key.

Who it does not fit

• You need a hosted service. There is no hosted offering, so you run the servers and you are the one on call.
• You need a Kubernetes replacement. There is no bin-packing scheduler and no CRDs, and autoscaling is beta.
• You need an edge network, an edge runtime or built-in image optimization. Those are not provided.
• You need a guaranteed memory ceiling. The figure above is a measurement on one box, not a limit, and no minimum server size has been benchmarked. A 1 vCPU, 1 GB server is enough to boot the control plane, and builds and several running apps are more comfortable with 2 vCPU and 2 GB, but treat both as starting points.

What is verified and what is not

Levelrail labels every feature by how well it is tested. See what is verified. A single node is the well-tested path and multi-node is labelled beta. Most stable-labelled features have a live end-to-end test on one machine, and almost none has a run on a fresh VPS with a real public domain.

The verified page lists each claim with its source and date, including what was checked on one host only and what has no evidence yet. Template numbers are a good example: 359 templates exist, 100 were smoke-tested and 93 started healthy, measured on 2026-10-25 on one arm64 Mac with a harness that emulates compose, not on a VPS.

How to check it yourself

Install it on a spare server, leave it idle, and read the Control plane health page, which shows live memory, CPU, goroutines and every background loop. Then deploy your real apps next to your current setup and compare. The install docs cover requirements, and the feature status page labels every feature as stable, beta or flag-gated.

If you are comparing platforms by size, read each project's own documentation for its figures and measure them on the same hardware as yours. The sourced comparison lists what each project states itself and marks everything else as not stated.

Next

• What is verified
• Coolify alternative
• Heroku alternative, self-hosted
• All alternatives
• Install
• Benchmarks in the docs