Dokploy alternatives: Coolify, Dokku, or a managed PaaS
The three real alternatives to Dokploy are Coolify (bigger community, more built-in services, ~800 MB idle), Dokku (Git-push deploys, no web UI, MIT-licensed) and paying a managed PaaS like Railway or Heroku. All three self-hosted options are free software; the cost is the server underneath, which is $17/mo for a 4 GB VPS (liquidweb.com/vps-hosting/self-managed/, checked 2026-09-09).
What this page is not: a Kubernetes comparison. None of these tools is a K8s distribution, and if you already run clusters, none of them will feel like an upgrade.
Which one should you pick?
| If you… | Pick | Why |
|---|---|---|
| Want the most services and templates out of the box | Coolify | 61,614 stars, 280+ one-click services, Apache-2.0 throughout |
| Want the lightest footprint and the cleaner UI | Dokploy | ~200 MB less idle RAM, but read the license section below first |
| Are happy in a terminal and want zero web UI | Dokku | MIT, 32,125 stars, runs on 1 GB of RAM |
| Deploy fewer than three small apps and hate sysadmin work | Railway or Heroku | $5–$21/mo and nothing to patch |
Star counts and licenses read from the GitHub API on 2026-09-09.
If you are choosing between Coolify and Dokploy specifically, the deciding factor is usually the license, not the feature list. See Is Dokploy actually open source? below.
How do the three self-hosted options compare?
| Coolify | Dokploy | Dokku | |
|---|---|---|---|
| GitHub stars (2026-09-09) | 61,614 | 37,191 | 32,125 |
| License (GitHub API) | Apache-2.0 | NOASSERTION — Apache-2.0 core plus a DSAL v1.0 carve-out | MIT |
| Latest release | v4.3.18, 2026-09-08 | v0.30.6, 2026-09-08 | v0.38.27, 2026-08-12 |
| Documented minimum | 2 cores, 2 GB RAM, 30 GB disk | 2 GB RAM, 30 GB disk (no CPU figure published) | 1 GB RAM with the Docker scheduler; 2 GB per node with k3s |
| Web UI | Yes | Yes | No — CLI and git push |
| Idle RAM we measured | ~800 MB | ~600 MB | Not measured |
| Multi-server | Documented, and Coolify's own docs call it "an experimental feature" | Remote Servers and Docker Swarm nodes, rated "High" setup complexity in Dokploy's docs | k3s scheduler |
Minimums come from each project's own install docs (Coolify, Dokploy, Dokku), read 2026-09-09. The RAM figures are ours: measured 2026-05-02 on local Docker with 4 vCPU and 8 GB, and not re-run against the current releases.
CapRover is the fourth name that comes up. It has 15,153 stars and was last pushed 2026-09-07, so it is alive, but GitHub reports its license as NOASSERTION and we have not run it. We are not recommending what we have not deployed.
What does each actually cost per month?
Same workload for all five columns: three small web apps and one PostgreSQL database, running around the clock.
| Dokploy / Coolify / Dokku on a VPS | Coolify Cloud | Dokploy Cloud | Railway Hobby | Heroku | |
|---|---|---|---|---|---|
| Software | $0 | $5/mo, up to 2 servers | $4.50/mo per server | Included | Included |
| Server | $17/mo (Liquid Web 4 GB Self-Managed) | Your own server, extra | Your own server, extra | Included | Included |
| Database | Self-hosted, no extra charge | Self-hosted, no extra charge | Self-hosted, no extra charge | Metered | Add-on, priced separately |
| Realistic monthly total | $17 | $22 | $21.50 | $5 credit, then metered | $21 (3 × $7 Basic dynos) |
Prices read on 2026-09-09 from liquidweb.com/vps-hosting/self-managed, coolify.io/pricing, dokploy.com/pricing, railway.com/pricing and heroku.com/pricing.
Two footnotes matter more than the totals.
Heroku's $5 Eco plan will not run three always-on apps. Eco buys 1,000 dyno hours per month, and Heroku's own docs say "this pool is shared by all Eco dynos in your account." Three dynos running continuously need about 2,232 hours. When the pool runs out, "all Eco dynos on your account are forced to sleep for the rest of the month" and you cannot buy more (devcenter.heroku.com, read 2026-09-09). Basic dynos at $7 each are the honest comparison.
The $17 VPS is the 4 GB tier, not the cheapest one. Liquid Web's Self-Managed line starts at $5/mo, but that is 1 vCPU and 1 GB, which is under Coolify's and Dokploy's own documented 2 GB minimum before you deploy a single app. The 4 GB tier is $17/mo regular and $8.50/mo for the first two months. Quoting the $5 tier against Railway would flatter us and mislead you.
If you run one small app and nothing else, Railway's Hobby plan is cheaper than any of this and you should take it.
Is Dokploy actually open source?
Partly. Dokploy's LICENSE.MD says content outside certain directories is Apache-2.0, but that anything under a /proprietary directory falls under a separate file. That file is the Dokploy Source Available License (DSAL) v1.0, and it reads: "This software and associated documentation files (the 'Software') may only be used in production, if you (and any entity that you represent) have agreed to, and are in compliance with, a valid commercial agreement from Dokploy." Development and testing are explicitly free.
This is why the GitHub API reports Dokploy's license as NOASSERTION rather than Apache-2.0. Coolify reports Apache-2.0 and Dokku reports MIT.
We could not find a top-level /proprietary directory on the canary branch on 2026-09-09, so the clause may currently be dormant. It is still the license you accept when you clone the repository, and it can stop being dormant in any release. Verify it yourself at github.com/Dokploy/dokploy/blob/canary/LICENSE.MD.
If your employer's legal review rejects source-available terms, that decides it: use Coolify or Dokku.
Can any of these run more than one server?
Yes, and all of them warn you about it in their own documentation.
Coolify's multi-server page opens with "This is an experimental feature," notes that servers must share the same architecture and images must go through a Docker registry, and does not automate the load balancer — you configure that yourself (coolify.io/docs, read 2026-09-09). Its Docker Swarm page carries the same experimental banner.
Dokploy documents three options: single server, Remote Servers over SSH, and Swarm Nodes. Its own comparison table rates Swarm setup complexity as "High," and the page states plainly that Swarm brings "more setup and operational complexity, and Dokploy does not perform automatic storage cleanup on the extra nodes" (docs.dokploy.com, read 2026-09-09).
Neither project claims a production-grade, hands-off cluster. If high availability is a hard requirement rather than a nice-to-have, run k3s or a managed Kubernetes service and skip this whole category.
When should you not self-host a PaaS at all?
- You have one app that rarely changes. A single
docker-compose.ymland Caddy will do the same job with no dashboard to patch and no extra 600–800 MB of RAM. - Nobody on the team wants to be on call for a VPS. Self-hosting moves kernel updates, disk pressure and backups onto you. Railway and Heroku charge a premium precisely to take that away, and for two or three small apps the premium is a few dollars a month.
- You need SAML or SSO. Neither Coolify's nor Dokploy's open-source edition covers enterprise identity. Check the paid tiers before you build around them.
- Uptime is contractual. Both are single-server by default. If the VPS dies, everything on it dies with it.
Where to run whichever one you pick
All three tools install on a plain Ubuntu or Debian box with root SSH. The spec that matters is RAM: budget the PaaS overhead first, then your apps.
Liquid Web Self-Managed VPS — the 4 GB tier at $17/mo (2 vCPU, 4 GB RAM, 80 GB SSD, 3 TB transfer; $8.50/mo for the first two months, checked 2026-09-09). That leaves roughly 3.2 GB for your applications after Coolify's overhead, or 3.4 GB after Dokploy's. View Self-Managed VPS plans →
The obvious objection: Liquid Web's Managed VPS starts at $72/mo for the same 4 GB, and the 8 GB tier is $100/mo. Managed buys you 24/7 support, patching and firewall setup. Skip it here. Coolify and Dokploy already automate the parts a managed host would handle, so you are paying twice for the same work. Take Managed only if the server will host things outside the PaaS that you do not want to administer.
Once the PaaS is up, Plausible and Outline both deploy on it straight from their own compose files, and the rest of the self-hosted guides follow the same pattern.
For most teams, Coolify. It has 61,614 GitHub stars to Dokploy's 37,191 (GitHub API, 2026-09-09), ships 280+ one-click services, and is Apache-2.0 across the whole repository, whereas Dokploy's LICENSE.MD carves out a /proprietary directory under a source-available license that requires a commercial agreement for production use. Dokploy wins on footprint: we measured roughly 600 MB idle against Coolify's 800 MB on 2026-05-02. If RAM is your binding constraint, that 200 MB is real.
No. Coolify's install docs require 2 cores, 2 GB of RAM and 30 GB of disk; Dokploy's require 2 GB of RAM and 30 GB of disk. Both figures are for the platform alone, before you deploy anything. On a 1 GB server you would be swapping during your first build. Dokku is the exception: its docs list 1 GB of system memory as the minimum for the Docker scheduler, and it has no web UI to keep resident.
Yes, if you like the Heroku workflow. Dokku is MIT-licensed with 32,125 stars and shipped v0.38.27 on 2026-08-12, so it is actively maintained. It deploys with git push instead of a dashboard, runs on 1 GB, and has the smallest surface area of the three. The trade-off is that everything is a CLI command: no click-to-provision databases, no built-in service catalogue, and a steeper first hour.
The software is free and the server is $17/mo for a 4 GB Liquid Web Self-Managed VPS (checked 2026-09-09). Against three Heroku Basic dynos at $7 each, that saves about $4/mo, which is not the reason to switch. The reasons that hold up are unmetered builds, self-hosted databases with no add-on pricing, and no per-app cost as you add projects. If you have fewer than three apps, staying on a managed platform is the rational call.
Yes, because all three run standard Docker containers. Export the docker-compose.yml from the source platform, import it on the target, dump and restore any database with pg_dump or mysqldump, and re-enter environment variables by hand. Your application code never leaves your Git repository, so the lock-in is limited to the platform configuration you would rewrite in an afternoon.