Northflank vs Owning the Stack: What the 'Cloud-Native Coolify Alternative' Really Costs

Northflank calls itself the number-one Coolify alternative: all of Coolify's deploy-from-git convenience, without the single-box Docker setup, running instead on Kubernetes with managed CI/CD, hosting, and databases in one bundle. It is a good pitch. It also has a price, and the price is the part nobody puts in the headline.

Here it is up front: a typical small production stack — two web services, a worker, Postgres, Redis, and 500 GB of egress — costs about $146 a month on Northflank Cloud and about $33 a month as a minimal Cluster-API-managed fleet on Hetzner Cloud. Same shape, same always-on resources, 4.4x apart. The rest of this post shows the math line by line, where the lines cross as you grow, and what the markup genuinely buys — because some of it is worth paying for.

The pitch: Coolify's convenience without the single box​

Coolify won its audience by being free, open-source, and installable on any VPS: point it at a server, connect GitHub, and you have a personal Heroku. Its ceiling is architectural. It orchestrates Docker on machines you operate yourself — OS patches, SSH, firewalls, database backups, and recovery are your job, and multi-server setups are coordinated rather than truly fleet-managed. Even Coolify Cloud, at $5 a month for two connected servers plus $3 per additional server, only manages the dashboard; the workload servers and everything on them stay yours to run.

Northflank's answer is to keep the all-in-one experience and move it onto Kubernetes. One platform covers Git-connected CI/CD with buildpacks or Dockerfiles, preview environments per branch, managed database addons (PostgreSQL, MySQL, MongoDB, Redis, RabbitMQ) with HA, backups, and forking, plus logs, metrics, autoscaling, and secrets. There are three ways to consume it: Northflank Cloud (their infrastructure, usage-priced), BYOC (their control plane driving Kubernetes in your AWS, GCP, Azure, Oracle, or Civo account, with a platform fee on top of your cloud bill), and Enterprise for everything custom. A free Sandbox with two services, one database, and two cron jobs gets you in the door.

So the honest question is not whether Northflank is more capable than Coolify — it is — but what the managed convenience costs against owning the same stack, and where "cloud-native, but still someone else's control plane" differs from a platform you run and can modify yourself.

Pricing the same stack four ways​

Take one concrete, typical small-production stack and price it four ways with published prices only:

• 2 web services at 1 vCPU / 2 GB each
• 1 background worker at 1 vCPU / 2 GB
• 1 Postgres at 1 vCPU / 2 GB with 50 GB SSD
• 1 Redis at 0.5 vCPU / 1 GB
• 500 GB monthly egress

Northflank Cloud. Compute is $0.01667 per vCPU-hour and $0.00833 per GB-hour, sold as fixed plans. Four containers map to nf-compute-100-2 at $24 each ($96), Redis maps to nf-compute-50 at $12, storage is 50 GB at $0.15 per GB-month ($7.50), and egress is 500 GB at $0.06 per GB ($30). Total: about $146 a month. No seat fees, billed per second — that part of the pitch is real.

Northflank BYOC on Civo. The same footprint needs roughly 4.5 vCPUs and 9 GB of RAM, which fits on three Civo Medium instances (2 cores, 4 GB, 50 GB NVMe) at $21.73 each: $65.19. Northflank's BYOC platform fee is $0.01389 per vCPU-hour and $0.00139 per GB-hour, which on the requested capacity works out to $45.63 plus $9.13 — about $55. Civo includes unlimited data transfer, so egress is $0. Total: about $120 a month, split between two bills.

Coolify Cloud plus a Hetzner dedicated box. Coolify Cloud is $5 a month for the dashboard covering the one connected server. An AX41-class dedicated machine (6-core Ryzen, 64 GB RAM, 2x512 GB NVMe) runs roughly €57 a month, or about $62. Egress on dedicated Hetzner hardware is effectively a non-issue at this volume. Total: about $67 a month — but every line Northflank manages (database HA, backups, preview environments, OS patching) is now your runbook.

A minimal CAPI-managed Hetzner Cloud fleet. Note this is a real multi-node Cluster API fleet, not a single box doing duty as a strawman: One control-plane node and two workers on CX32 instances (4 shared vCPUs, 8 GB each) at €7.49 apiece is €22.47; a load balancer and a 50 GB volume add roughly €8 more. That is about €30, or roughly $33 a month, with 20 TB of included traffic per node making the 500 GB of egress free.

Two things stand out. First, the BYOC platform fee alone ($55) costs more than the entire owned fleet ($33) — "run it in your own cloud" still meters every vCPU through someone else's cash register. Second, egress is a quiet third of the Northflank Cloud bill: $30 of the $146 is bandwidth that flat-rate providers throw in.

Where the lines cross​

One data point invites cherry-picking charges, so here is the same comparison at three sizes:

• Side project (one small service, one tiny database, 50 GB egress): Northflank Cloud lands around $15 a month (small plans plus a few dollars of storage and egress). The owned equivalent is a single CX22 cloud VM at about €4.35 — roughly $5. Managed wins on convenience here, and the cash gap is ten dollars, not a hundred.
• Typical small production (the stack above): $146 vs $33. The 4.4x gap is the steady-state story for always-on workloads.
• Growth (ten services, a couple of workers, a larger database): Northflank Cloud scales per container toward $400–500 a month, while the owned fleet absorbs it with two or three more CX32 nodes — roughly $50–60 a month total. The gap widens past 7x because per-second billing meters every container while flat-rate hardware just fills up.

The pattern is simple: managed per-unit pricing wins when workloads are small or idle enough to scale to near-zero, and flat-rate owned hardware wins the moment the stack is always on — which is precisely what "production" means. The crossover for a typical team sits somewhere around two or three always-on services with a database: below that, the $10–20 managed bill is cheaper than your time; above it, you are renting the same silicon back at a steep multiple.

What the markup actually buys​

None of the above means the $146 is a scam. It buys specific, real things, each of which maps to work you would otherwise do:

• Managed databases with HA, backups, and forking. One-click Postgres with replicas, point-in-time recovery, and branch-style forks. The DIY equivalent is operating Postgres yourself — Patroni or CloudNativePG, backup cron jobs, restore drills — which is a genuine skill and a genuine pager.
• Per-branch preview environments. Full-stack ephemeral environments (services plus databases plus URLs) spun up automatically per pull request. Reproducible on your own fleet with preview namespaces and automation, but nobody does it on day one.
• Zero Kubernetes operations. No cluster upgrades, no CNI/CSI debugging, no control-plane babysitting. A CAPI fleet is declarative and self-healing, but somebody still owns the management cluster and the upgrade runbook.
• Per-second billing with no seats. Teams are free, idle resources scale down, and there is no $399-a-month platform floor like some competitors charge. For spiky or experimental workloads this is structurally cheaper than reserved hardware.
• Compliance posture. SOC 2 Type II out of the box, which matters the first time an enterprise prospect sends a security questionnaire.

Priced against an engineer's hourly rate, one avoided database-restore incident can pay for a year of the markup. The mistake is treating that as a permanent verdict rather than a stage: what is worth $113 a month in avoided ops at three services is worth a lot less per unit at thirty, when you have a platform engineer anyway.

The cost beyond the bill: someone else's control plane​

Money is only the metered part. The structural difference is who owns the control plane:

• BYOC does not end the metering; it moves it. Your data plane runs in your cloud account, but the platform fee ($0.01389/vCPU-hour, $0.00139/GB-hour) applies to capacity you already pay your cloud provider for. It is two margins on the same core.
• Your favorite cheap cloud may not be invited. Self-serve BYOC covers AWS, GCP, Azure, Oracle, and Civo — Hetzner is not on the list, and bare-metal/on-premises BYOK is an eligibility conversation, not a signup button. The cheapest silicon is reachable only through the most expensive sales motion.
• You cannot modify the platform itself. A deploy pipeline quirk, a missing database extension, a routing behavior you disagree with — all of these are feature requests, not pull requests. On an open-source stack you own, they are Tuesday.
• Residency is split. With BYOC your workloads sit in your region and account, but the control plane — the thing that can read your configs, logs, and secrets metadata — is still someone else's SaaS. For regulated or paranoid teams that distinction does real work in a threat model.

This is the exact seam that matters: "cloud-native, but still someone else's control plane" is a meaningfully different pitch from a self-hosted platform a team can run and modify itself. Northflank abstracts the machines away; owning the fleet means there is no abstraction between you and the machines at all.

Verdict: who should pick which​

• Pick Northflank Cloud if you are pre-platform-team, your workloads are small or spiky, and $100–150 a month is less than the ops hours you would burn. Side projects, prototypes, and early startups live here happily.
• Pick Northflank BYOC if you need the managed developer experience but workloads must run in your own cloud account for credits, commits, or data-residency reasons — and you have priced the platform fee on top rather than discovering it later.
• Pick Coolify on a dedicated box if you want minimal spend with a friendly UI and accept single-server operations: fine for agencies and side businesses, outgrown when you need a second machine to behave like the first.
• Own a CAPI-managed fleet when the stack is always on, growing, and boring enough to deserve flat-rate hardware — typically around three-plus services with a database. The $33 fleet above is not a heroic minimum; it is one control-plane node, two workers, and a load balancer.

The through-line: start where the ops burden matches your team, and re-check the math every time the bill doubles. Convenience compounds like interest — in both directions.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.