Coolify Cloud charges $5/month for two managed servers. The Hetzner CX22 it most often points at costs €4.35/month with 20 TB of traffic included. Both numbers are real, and neither one is the whole bill.
If you run one hobby app on one Hetzner box with Coolify, you pay roughly €4.35 + $2.50 per server for the control plane — the cheapest managed PaaS subscription you can get that still leaves you owning the hardware. If you run eight boxes, you pay €4.35 × 8 + $29. The box price scales linearly. The platform fee does not — and the gap between what the $5 rents and what the box still ships with you is exactly where the single-box self-hoster and a fleet platform diverge.
Here is the answer up front, with the numbers this teardown is built on. All Hetzner prices are the CX-class shared-vCPU line (2 vCPU, 4 GB RAM, 40 GB SSD, 20 TB traffic). Coolify Cloud is the managed dashboard: $5/month for 2 servers, then $3 per extra server. You still bring — and pay for — your own VPS in both self-hosted and Cloud modes.
| Fleet size | Raw Hetzner only (CX22-class) | Hetzner + Coolify Cloud | Effective platform cost per server | Hetzner + flat fleet platform (no per-server rent) |
|---|---|---|---|---|
| 1 server | €4.35/mo (~$4.70) | €4.35 + $5.00 = ~$9.70 total | $5.00/server | €4.35 + $0 per-server rent = ~$4.70 |
| 2 servers | €8.70/mo (~$9.40) | €8.70 + $5.00 = ~$14.40 | $2.50/server | ~$9.40 |
| 5 servers | €21.75/mo (~$23.50) | €21.75 + $14.00 = ~$37.50 | $2.80/server | ~$23.50 |
| 8 servers | €34.80/mo (~$37.60) | €34.80 + $23.00 = ~$60.60 | $2.88/server | ~$37.60 |
| 10 servers | €43.50/mo (~$47.00) | €43.50 + $29.00 = ~$76.00 | $2.90/server | ~$47.00 |
Two footnotes the headline price hides: CX22 at €4.35 is the pre-April 2026 list price you will still see quoted in tutorials. Hetzner raised cloud servers 30–37% on April 1, 2026, then added another ~33% on June 15 for the CX/CAX shared-vCPU lines (113–175% on the CPX/CCX dedicated lines). The CX22 generation itself was retired and replaced by CX23; the entry tier now lists at €5.49/month ex-VAT (~€6.59 with VAT, ~$7.10). We keep €4.35 in the table because the comparison is about the shape of the fee, not the exact euro — but your real invoice today is closer to €5.49–€6.59 per box, so inflate every Hetzner column by ~30% before you budget.
The shape matters more than the exact euro: Coolify Cloud's fee is flat for two, then linear per server after that. A fleet platform that bundles machine lifecycle into the control plane has no per-server rent at all — you pay for boxes, the platform scales with them. The rest of this post is what each line actually buys, and where that shape breaks.
What the $5 Actually Buys — and What It Decidedly Does Not
Coolify is an open-source, Apache-2.0 self-hosted PaaS with roughly 54,000–57,000 GitHub stars and 575 contributors as of mid-2026. It wraps Docker with a dashboard, git-push deploys, automatic TLS via Traefik/Let's Encrypt, one-click databases, Docker Compose support, and a catalog of 280+ one-click services. The current stable release is v4.0.0 (April 27, 2026) after a two-year run as v4.0.0-beta.*, with v5 already in early development.
You have two ways to run it:
- Self-hosted (free): You install Coolify on a VPS you own. The dashboard runs on that VPS. You manage the Coolify installation itself — pulling updates, restarting Docker, backing up Coolify's own database.
- Coolify Cloud ($5/month for 2 servers, $3 per extra): You sign up at
app.coolify.io. Coolify hosts the dashboard. You connect your own servers over SSH. Your apps still run on your servers — Coolify Cloud is a managed control plane, not app hosting.
That distinction is the entire pricing story. MapleDeploy's comparison puts it cleanly: Coolify Cloud manages the Coolify instance, not your server. The operating system, the firewall, the disk, the security updates on the machine your apps run on — all still yours.
Concretely, the $5 buys:
- Hosted dashboard and control plane — no Coolify container to run, backup, or upgrade on your own box. High availability of the UI itself is Coolify's problem, not yours.
- Managed Coolify updates — Coolify checks for updates and shows them in the dashboard. In Cloud mode the platform handles the rollout. In self-hosted mode you click Update yourself; Docker-related updates restart Docker, which restarts every app until it comes back.
- Support and account management — team access, connection management for multiple servers from one pane.
What the $5 does not buy, and what the name Cloud can easily imply it does:
- No server provisioning. You create the Hetzner (or AWS, GCP, other) server yourself — via the Hetzner console, API, or Terraform.
- No OS patching. Server patching in Coolify (available since
v4.0.0-beta.419) lists available OS packages and lets you click Update or Update All Packages. It never auto-installs.Dockeror kernel updates will still bounce the box. - No firewall, disk, backup, or network management beyond what the Coolify agent configures for its own reverse proxy. Backups for Coolify's own state are on you in self-hosted mode; app data volumes are yours in both modes.
- No compute, storage, or bandwidth. The $5 does not include a single byte of egress or a single hour of vCPU. Your Hetzner bill is separate and unchanged.
At $2.50 per server for the first two, it is genuinely cheap for what it is — the least expensive managed-PaaS control plane you can buy that does not also tax you per seat or per GB. The question is not whether $5 is too much for a dashboard. It is what happens when per-server meets fleet.
What €4.35 Bought — and What €5.83 Buys Today
The Hetzner CX22 at €4.35 is the anchor every self-hosting tutorial quoted from 2023 through early 2026 for good reason. For that price you got 2 shared Intel vCPUs, 4 GB RAM, 40 GB local SSD, and 20 TB of included traffic on a box that provisions in under a minute with a clean API. The 20 TB number is doing heavy lifting — compare it to the metered egress on any hosted PaaS at $0.05–$0.15/GB, where 20 TB would add $1,000–$3,000 before compute even enters the chat.
That price did not survive 2026.
- April 1, 2026: Hetzner raised cloud servers 30–37% across regions, object storage 30–53%, and memory add-ons by roughly 575%, citing "dramatically increased costs to operate infrastructure and buy new hardware" — the same DRAM/HBM squeeze driven by AI accelerator demand that this blog has tracked as a memory supply-chain story all year.
- June 15, 2026, 08:00 CEST: A second, uneven adjustment landed only on new orders and rescales (existing instances kept their old price until a rescale forces the new list). Shared-vCPU CX/CAX lines rose another ~33%. Dedicated-vCPU CPX/CCX lines jumped 113–175% — CPX22 +144%, CCX13 +169%, CCX23 +173%.
- Generation rotation:
CX22/CX32/CX42stopped being orderable. They were replaced by CX23/CX33/CX43 with the new list price baked in. Any tutorial still quoting "Hetzner CX22 from €3.79" is naming a plan you cannot buy at a price that no longer applies.
The practical number today, ex-VAT, is €5.49/month for the entry CX-class box in the new generation; with German 19% VAT about €6.53–€6.59. We keep the teardown anchored at €4.35 so its shape stays comparable to the $5 Coolify line every reader already has in their head, but add ~30% to every row if you are provisioning fresh this quarter.
That still leaves the bundle absurdly cheap for what you get — one invoice line, no per-GB meter, 20 TB you would pay four figures for elsewhere already included. What it also leaves is every operational question the box never promised to answer.
Line by Line: 1, 3, and 8 Servers
A single price comparison cannot answer a PaaS question. Three worked fleets can — one that fits on one box, one that outgrows it, and one that forces a real provisioning strategy. Each fleet assumes Hetzner CX-class boxes at the tutorial-era €4.35 (inflate ~30% for fresh orders), Coolify Cloud at $5 + $3/extra, and the included 20 TB already covering all three fleets' bandwidth at this scale.
Fleet A: One App, One Box
A side project, a Next.js app with Postgres and Redis, average 30 GB egress. All fits on one CX22.
| Line | Monthly cost |
|---|---|
| Hetzner CX22 | €4.35 (~$4.70) |
| Coolify self-hosted | $0 (you update it) |
| Coolify Cloud (uses 1 of 2 slots) | $5.00 flat — $5.00 per used server, $2.50 per included slot |
| Total self-hosted | ~$4.70 |
| Total with Cloud | ~$9.70 |
| Bandwidth overage | $0 — 20 TB included covers 30 GB 666× over |
At one server, Coolify Cloud more than doubles the bill for the convenience of not running the dashboard yourself. That is a fair trade if you never want to SSH in to docker compose pull && docker compose up for Coolify's own updates, but it is not a compute saving. Self-hosting Coolify here is literally $0 of platform fee.
Fleet B: Three Apps, Three Boxes (or One Box Pushed Hard)
A small team: marketing site, API + Postgres, and a worker with a managed queue. Each wants isolation or its own deploys. Three CX22s, about 200 GB total egress.
| Line | Monthly cost |
|---|---|
| Hetzner 3× CX22 | €13.05 (~$14.10) |
| Coolify Cloud (needs 1 extra beyond 2) | $5 + $3 = $8.00 |
| Total with Cloud | ~$22.10 |
| Effective Cloud cost per server | $2.67 |
| Bandwidth overage | $0 |
The marginal third server costs €4.35 for the box plus $3.00 for the platform — $7.70 for one more placement. At this size you also start manually wiring cross-server concerns: which box runs which service, what happens when the Postgres box needs more RAM than a CX22 holds, how deploys target the right host.
Fleet C: Eight Apps, Eight Boxes, One Team
A team that outgrew "one box per app" without ever choosing a fleet strategy. Eight CX22s, a mix of sizes, about 1 TB egress. This is where the per-server rent line finally reads as a tax.
| Line | Monthly cost |
|---|---|
| Hetzner 8× CX22 | €34.80 (~$37.60) |
| Coolify Cloud (6 extras beyond 2) | $5 + (6 × $3) = $23.00 |
| Total with Cloud | ~$60.60 |
| Platform fee as share of total | 38% |
| Bandwidth overage | $0 |
At eight servers, Coolify Cloud adds the equivalent of five more raw boxes just in platform rent. The boxes themselves are still cheap. The per-server shape is where the bill concentrates — not because $3 is large, but because it is the only line that scales with every machine you add.
A fleet platform with no per-server rent (the model bex uses) flattens that line to $0. The same eight boxes cost ~$37.60 and the platform scales with them, because provisioning is the platform rather than a subscription layered on top of each placement. The 38% delta at fleet C is not Coolify being expensive — at $2.88 per server it is still the cheapest managed PaaS control plane you can buy. It is the structural difference between renting a dashboard per placement and owning a fleet.
Where Coolify Stops and a Fleet Begins
Coolify, Dokploy, CapRover, and the current wave of self-hosting tools all solve the same first problem — manage the box you already have — and solve it well. Dokploy leans on Docker Swarm with Traefik, CapRover on a veteran single-server model, Coolify on a richer Compose and service catalog (280+ one-clicks). By 2026, Deploynix's showdown could fairly call Coolify "the most feature-complete all-rounder" with native multi-server support, and log-traces of Coolify v4 users do run multiple servers behind one dashboard.
The seam is not whether one dashboard can see two servers. It is what happens when provisioning, placement, and lifecycle need to be declarative and fleet-wide rather than per-host.
Four things concretely break when a team that adopted a single-box tool for its first app needs a second, third, and eighth machine:
1. No declarative machine lifecycle. Hetzner servers for Coolify are created outside Coolify. Coolify connects to them over SSH and configures them with its agent. A Cluster API provider for Hetzner (CAPH) does the opposite: machines are Kubernetes custom resources. A MachineDeployment says "three CX33 nodes in fsn1 with label workload=general," and the provider reconciles it — creating, draining, replacing, and rolling nodes without an operator opening the Hetzner console. Coolify's own docs keep server creation and lifecycle firmly on the operator.
2. No fleet-wide placement. Coolify lets you pick a server per service at deploy time. A Kubernetes scheduler — and a Cluster API fleet behind it — places workloads across the fleet by resource fit, affinity, taints, and autoscaler signals. When one box fills up, the next workload lands elsewhere without a human choosing the destination.
3. Health, rollout, and scale are Docker primitives, not fleet primitives. Coolify deploys are Docker container updates on a host. Health checks, graceful rolling updates, autoscaling, and PodDisruptionBudgets are Kubernetes-level contracts that survive a host going away. Orkai's retrospective is blunt: "We used Coolify, Dokploy, CapRover — they're great projects ... but when we needed real health checks, graceful rolling updates, autoscaling, or the ability to grow from one node to three, we found ourselves fighting the abstraction instead of shipping."
4. The update seam reappears at the OS layer. Coolify patches itself. Your eight servers still need OS patching, kernel reboots, Docker restarts, volume backups, and a strategy for the box that dies at 2 a.m. Coolify surfaces available packages; it never auto-installs them, for good reason — a Docker restart bounces every app on the host. A Cluster API fleet encodes that operational knowledge as machine images, rolling update strategies, and node replacement rather than per-server button clicks.
None of this makes Coolify the wrong choice — for one or two servers it is the cheapest, fastest path to a Vercel-like deploy experience you own, running on the cheapest raw compute you can buy. The seam is that the cheapest path for one placement stays a per-placement path at eight. A fleet platform stops charging per placement precisely because the work that breaks at eight is the platform's own job.
When Coolify Wins, When a Fleet Wins
- One box, one hobby app, solo operator — Coolify self-hosted wins. Run Coolify on the same CX22 (or a CX23) that runs your app. Platform fee $0. Your time is the update click. No second machine to provision, no scheduler to learn.
- Two to three boxes, small team, no autoscaler — Coolify Cloud probably wins. At $5 for two ($2.50 per box) and $8 for three ($2.67 per box), the managed dashboard pays for itself in the first hour you do not spend babysitting Coolify's own install. Stay here as long as you never ask "where should this new service land?"
- Second machine that matters (different region, team isolation, scaling) — fleet starts winning. As soon as you need declarative provisioning ("give me three of those in
nbg1"), fleet-wide scheduling, or a rolling strategy that survives a host loss, the per-server platform rent is paying for a dashboard while you still own the fleet operations. That is the exact seam a Cluster API–managed fleet is built to not have — the Hetzner bill is the only bill that scales with machines, because the fleet is the abstraction.
The honest cost of the self-hosted path at this boundary is operator time. How long does it take your team to provision, patch, place, and recover eight machines with the tooling you chose? At one box that time is an hour. At eight it is an on-call rotation — whether it is visible as a platform fee or not.
The Bottom Line
At roughly 57,000 stars and 575 contributors, Coolify is the most proven single-box self-hosting PaaS you can adopt — and at $5 for two servers it is the cheapest managed dashboard that still lets you own the underlying Hetzner boxes and their 20 TB of included bandwidth. No hosted PaaS touches the raw-compute value of a €4.35–€6.59 CX-class box for a bandwidth-heavy workload; the metered alternatives would charge four figures for the same traffic.
What the $5 covers is precisely scoped: a hosted dashboard, managed Coolify updates, and support — not server provisioning, not OS patching, not the VPS itself. What the per-server shape costs is also precisely scoped: $2.50 per box for two, $3.00 per marginal box after that, compounding into 38% of the total at eight nodes. Self-hosting Coolify itself erases that line, but trades it for the operator hours the dashboard was there to save.
A fleet platform makes a different trade. By making Hetzner machines declarative — provisioned, placed, and rolled by the same control plane that runs workloads — it eliminates the per-server dashboard rent entirely. You still pay for boxes. You stop paying per box for the platform to see them.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. A Cluster API–managed fleet, a Render-compatible API, and an MCP server so AI agents can operate the platform as first-class deployers, not just code generators. Star the repo on GitHub or deploy your first app today.