# Lightest Self-Hosted PaaS Options by RAM (2026)

> Levelrail, CapRover, Dokku, Coolify and Dokploy ranked by the RAM each project's docs state, next to a measured idle figure, for small and cheap VPS plans.

Canonical URL: https://levelrail.com/blog/lightest-self-hosted-paas/

> [!TIP]
> Levelrail is open source (Apache 2.0). Read the code, star it and try it on GitHub: [github.com/glincker/levelrail](https://github.com/glincker/levelrail), or [install it in one command](https://levelrail.com/install/).

If you are buying a small VPS, the short answer is this: Levelrail, CapRover and Dokku all state a 1 GB minimum, and Coolify and Dokploy state 2 GB. Levelrail is first in this list, and it is our product, so weigh that. Our reason is that it keeps a dashboard, metrics, logs and rollback inside a control plane that measured about 200 MB at idle, where the others lean on extra tools for some of that.

Two kinds of numbers appear below, and they should not be mixed up. The ranking uses the minimum RAM each project states in its own documentation. Separately, we report one measured idle figure for Levelrail. A stated minimum is a recommendation. A measured idle figure is an observation. We have no matching measurement for the other four, so we do not put them in one column.

The facts about other projects come from [our comparison data](https://levelrail.com/compare/), read on 2026-11-18, and each one links to the official page it came from. Where a project's docs do not say something, we leave it out. For the wider field, see [the alternatives page](https://levelrail.com/alternatives/) and the earlier post on [self-hosted Vercel and Heroku alternatives](https://levelrail.com/blog/self-hosted-vercel-heroku-alternatives/). There is also a shorter reference page on [lightweight self-hosted PaaS](https://levelrail.com/lightweight-self-hosted-paas/).

## Quick pick table

| Rank | Name | Built around | Stated min RAM | License |
| --- | --- | --- | --- | --- |
| 1 | Levelrail | Docker apps with a built-in dashboard, metrics and rollback | 1 GB (about 200 MB idle, measured once) | Apache 2.0, one edition |
| 2 | Dokku | Heroku style `git push` on one host | 1 GB (Docker scheduler), 2 GB per node (k3s) | MIT, Dokku Pro is paid |
| 3 | CapRover | A Swarm based panel with a one-click app store | 1 GiB | Apache 2.0, some Pro features reserved |
| 4 | Coolify | A large service catalog over plain SSH | 2 GB | Apache 2.0 |
| 5 | Dokploy | A Docker Swarm based dashboard | 2 GB | Apache 2.0 plus a source-available folder |

Ties at 1 GB are broken by what each project documents about its running footprint, and that is a judgment call. The stated minimum for Levelrail is its [install page](https://levelrail.com/install/) starting size, 1 vCPU, 1 GB RAM and 10 GB disk, and that page says no minimum has been benchmarked or published yet. Footnote: the stated minimums are what the projects tell you to provision, and the 200 MB is a measured idle figure for a different thing, so they are different kinds of numbers.

## 1. Levelrail

Levelrail is a self-hosted platform that runs your apps in Docker from one Go control plane binary, with one agent per extra node. It is Apache 2.0, one edition, no paid tier ([license notes](https://docs.levelrail.com/license-and-notices)). Install it with one command from the [install page](https://levelrail.com/install/).

The memory story is a measurement, not a promise. The control plane at idle used about 200 MB in one measurement on one machine. The method and the raw numbers are on the [measured page](https://levelrail.com/measured/). Apps you deploy and any image builds are extra, so size the server for those too.

What it documents that the other four projects do not, row by row from the comparison data:

- Metrics at 15 second intervals and full-text log search are built in, with crashloop detection that shows the last log lines ([observability docs](https://docs.levelrail.com/observability)). Dokku's data row says to add your own stack, and Coolify, Dokploy and CapRover document partial coverage that points to external systems or optional add-ons for search.
- Rollback is labeled stable. Prior images are pinned so cleanup cannot remove them, and a 5xx spike or crashloop after cutover rolls back automatically, which has been live-tested. The latency signal is beta ([rollback docs](https://docs.levelrail.com/rollback)).
- Deploys cut over only after the readiness probe passes. This is live-tested on one node, and multi-node is beta ([deploying apps](https://docs.levelrail.com/deploying-apps)).
- Allow and Deny IAM policies (stable), passkeys, TOTP and an audit log with a verifiable hash chain are in the one edition ([access control](https://docs.levelrail.com/access-control)). Dokploy lists audit logs under its Enterprise license, and CapRover lists 2FA as a Pro feature.
- An opt-in public status page, labeled stable, with 90 day uptime bars and incidents ([status page](https://docs.levelrail.com/status-page)), and per-app cron jobs with run history and retries ([scheduled jobs](https://docs.levelrail.com/scheduled-jobs)).
- Remote nodes dial out over mTLS, so no inbound ports are needed on them. This control channel is labeled beta ([multi-node](https://docs.levelrail.com/multi-node)).

Templates are active and growing: the catalog has 592 entries, 583 active, and 346 of those are proven on a real engine, while 246 are previews without that pass yet. The label on that feature is beta ([templates status](https://docs.levelrail.com/templates-status)). Browse them on the [templates page](https://levelrail.com/templates/).

Moving from one of the others is covered by read-only importers: see [/migrate/coolify/](https://levelrail.com/migrate/coolify/), [/migrate/dokploy/](https://levelrail.com/migrate/dokploy/) and [/migrate/caprover/](https://levelrail.com/migrate/caprover/). There is no importer for Dokku. Gaps are listed on [feature status](https://docs.levelrail.com/feature-status): for example, automatic public TLS has one recorded run on a live VPS, and multi-node cross-host transport is still being verified.

Verdict: first on this list because it pairs a 1 GB stated floor with a measured idle control plane near 200 MB and keeps observability, rollback and access control in the same binary. Per-feature sources are on the pages for [Coolify](https://levelrail.com/compare/coolify/), [Dokploy](https://levelrail.com/compare/dokploy/), [CapRover](https://levelrail.com/compare/caprover/) and [Dokku](https://levelrail.com/compare/dokku/).

## 2. Dokku

Dokku is a Docker powered mini-Heroku. You drive it with commands on the host over SSH or `git push`, and the data lists it as hard to beat for simplicity on a single server. The comparison data notes a very small idle footprint, with no long-running control service beyond a small event listener ([architecture](https://dokku.com/docs/development/architecture/)). We have not measured it ourselves.

- Stated minimum: 1 GB with the Docker scheduler, 2 GB per node with the k3s scheduler, no CPU minimum stated ([installation](https://dokku.com/docs/getting-started/installation/)).
- License: MIT. Dokku Pro is a paid add-on, listed at $849 one-time (early bird) for 1 production and 2 pre-production servers ([Dokku Pro](https://dokku.com/docs/enterprise/pro/)).
- Latest release in our data: v0.38.34, 2026-10-30. The install docs list Ubuntu 22.04, 24.04, 26.04 and Debian 11+ on amd64 and arm64.
- Documented gaps against Levelrail: no built-in metrics ([our Dokku page](https://levelrail.com/compare/dokku/)), failed image deploys restore the previous image but there is no one-command rollback ([builder management](https://dokku.com/docs/deployment/builders/builder-management/)), and users are SSH keys with no role-based access ([user management](https://dokku.com/docs/deployment/user-management/)). Its JSON API is in the paid Pro add-on.

Fit: built around the Heroku workflow on a single server, driven from a terminal.

## 3. CapRover

CapRover is a Swarm-based platform with a web dashboard, a CLI and a one-click app store that accepts third-party template repositories. The current release in our data is v1.16.1, 2026-11-18.

- Stated minimum: 1 GiB of RAM and 5 GiB of free disk, no CPU minimum stated ([get started](https://caprover.com/docs/get-started.html)).
- License: Apache 2.0, with some Pro features reserved ([CapRover Pro](https://pro.caprover.com/)). The Pro page lists 2FA and login alerts, and no roles or audit log are documented.
- Metrics come from optional NetData, and app and build logs are viewable with no search documented ([NetData](https://caprover.com/docs/server/netdata)). Release v1.16.0 added runtime health, task history and disk alerts.
- Updates: stateless apps default to start-first with an optional health check, and apps with mounted data default to stop-first. The docs do not promise interruption-free updates ([scaling](https://caprover.com/docs/scaling)).
- Remote nodes are added with an SSH key from the dashboard ([multi-node](https://caprover.com/docs/scaling/multi-node)).

Fit: built around Docker Swarm with a simple panel and an app store. If you run it today, the [CapRover importer](https://levelrail.com/migrate/caprover/) reads your setup without changing it.

## 4. Coolify

Coolify is a dashboard that manages servers you already have over plain SSH, and it is known for its large catalog of one-click services. The current release in our data is v4.7.5, 2026-11-10.

- Stated minimum: 2 GB of RAM, 2 CPU cores and 10 GB of disk ([installation](https://coolify.io/docs/get-started/installation)).
- License: Apache 2.0, and the data records no paywalled features self-hosted. Cloud is $5 a month for the base plan with 2 servers.
- Architecture: three base containers, the app, PostgreSQL and Redis ([how Coolify works](https://coolify.io/docs/core/how-coolify-works)).
- Observability: Sentinel charts for CPU, memory, disk, network and load, with log drains to external services. The docs say log queries and dashboards need an external system ([monitoring](https://coolify.io/docs/core/observability/monitoring/overview)).
- Deploys start the replacement first, but the docs state there is no zero-downtime guarantee ([rolling updates](https://coolify.io/docs/applications/deployments/rolling-updates)).

Fit: built around a big service catalog and a polished dashboard on top of SSH. The [Coolify importer](https://levelrail.com/migrate/coolify/) supports `--dry-run`, so you can run both side by side before moving anything.

## 5. Dokploy

Dokploy is a Docker Swarm based platform with a modern dashboard, an official MCP server and S3 backups. The current release in our data is v0.31.4, 2026-11-17.

- Stated minimum: 2 GB of RAM and 30 GB of disk, no CPU minimum stated ([installation](https://docs.dokploy.com/docs/core/installation)).
- License: Apache 2.0 plus a source-available `/proprietary` folder. Self-hosted is free, and Cloud Hobby is $4.50 a month per server.
- Architecture: three parts, a Next.js app, PostgreSQL and Traefik ([architecture](https://docs.dokploy.com/docs/core/architecture)).
- Access: Owner, Admin and Member roles plus 2FA and passkeys are in self-hosted, while custom roles and audit logs need an Enterprise license ([enterprise](https://docs.dokploy.com/docs/core/enterprise)).
- Rollbacks: a Swarm health check with automatic rollback on failure is documented, and it needs a `/health` endpoint and curl in the image ([rollbacks](https://docs.dokploy.com/docs/core/applications/rollbacks)).

Fit: built around Docker Swarm with a fast release cadence. The [Dokploy importer](https://levelrail.com/migrate/dokploy/) is read-only and has `--dry-run` and `--only` flags.

## How to read the RAM column

A stated minimum tells you what the authors expect a working install to need, and projects set that bar differently. Some include room for builds, some count only the platform. Treat the ranking as a guide to the floor, then check each installation page linked above before you buy a plan.

Idle use also depends on what is running. The about 200 MB for Levelrail is one measurement of the control plane at idle, taken once, and the [measured page](https://levelrail.com/measured/) shows how. It is a useful signal for a small server and it is not a benchmark of the field.

Whatever you pick, leave headroom for your own apps and databases, since those will use more memory than the platform does. A [templates](https://levelrail.com/templates/) install or an image build can briefly use far more than idle. The [five-minute install guide](https://levelrail.com/blog/install-levelrail-five-minutes/) shows a first deploy end to end.

## Summary

For a 1 GB VPS, Levelrail, Dokku and CapRover state a floor that fits, and Coolify and Dokploy state 2 GB. Levelrail is first here because its measured idle control plane is about 200 MB and it keeps metrics, logs, rollback and access control in one binary. Read the [comparison pages](https://levelrail.com/compare/), try the [install](https://levelrail.com/install/), and check [feature status](https://docs.levelrail.com/feature-status) for what is stable and what is beta.

## Frequently asked questions

### What is the best lightweight self-hosted PaaS?

By stated minimum RAM, Levelrail, CapRover and Dokku all list about 1 GB (Levelrail 1 GB, CapRover 1 GiB, Dokku 1 GB with the Docker scheduler), while Coolify and Dokploy list 2 GB. Levelrail is our product and the one we rank first: its control plane measured about 200 MB at idle in one measurement, and it adds a dashboard, metrics, logs and rollback on top. The stated minimums and the measured figure are different kinds of numbers.

### How much RAM do Coolify and Dokploy need?

Coolify's installation guide lists 2 GB of RAM with 2 CPU cores and 10 GB of disk. Dokploy's installation guide lists 2 GB of RAM and 30 GB of disk, with no CPU minimum stated. Both numbers come from the projects' own docs as of 2026-11-18.

### Can I run a self-hosted PaaS on a 1 GB VPS?

The docs of Levelrail (1 vCPU, 1 GB, 10 GB), CapRover (1 GiB, 5 GiB free disk) and Dokku (1 GB with the Docker scheduler) state a 1 GB minimum. That is the floor for the platform itself. Your apps and any builds you run on the same server need memory on top of it, so a 2 GB server leaves more room.

### Is a stated minimum the same as idle memory use?

No. A stated minimum is what a project tells you to provision. Idle use is what the running software actually holds when nothing is happening. Levelrail's measured idle control plane of about 200 MB is one measurement on one machine (see the measured page), so do not read it as a minimum or as a like for like comparison with the other projects' stated figures.
