Best European Alternatives to Heroku

Why people leave Heroku

Heroku is not abandoned, whatever you have read. Fir, the Kubernetes-based generation, is generally available and Heroku recommends it for new applications and actively developed projects. The documentation is current. Somebody is clearly still working there.

What has happened is subtler and harder to plan around: Heroku has split in two, and the halves cost very different amounts. The cheap Heroku everyone remembers, Eco dynos at $5 and Basic at $7 in the Common Runtime, is Cedar, the legacy generation. Fir has no Common Runtime yet, so a Fir application lives in a Private Space, and Private Spaces are available only to verified Heroku Teams and Heroku Enterprise. The modern path starts at a procurement conversation.

The third thing is the one nobody argues about. Free dynos, free Postgres and free Data for Redis stopped on 28 November 2022. That is the date a generation of side projects, university courses and internal tools went looking for somewhere else, and it is why this page has a much shorter shortlist than it should: Europe has very few real platform-as-a-service products, and two of them are worth your time.

• The modern generation is the expensive one Fir is what Heroku recommends and what its product work is going into, and Heroku's own generations documentation lists what it does not have yet: Common Runtime, Shield compliance, Heroku CI, dyno autoscaling, VPC peering, VPN support, internal routing and Clojure. No Common Runtime means Private Spaces, and Private Spaces mean a verified Team or Enterprise. A solo developer who follows the recommended path arrives at a sales form.
• The EU region is one region with no address The Common Runtime offers exactly two regions, us and eu, and the documentation gives the eu region no stated location. The places you can actually name, Dublin and Frankfurt, are Private Spaces regions, which puts European data residency behind the same Team or Enterprise gate as everything else. If your reason for asking was a procurement rule, "eu" is a weak answer to put in the box.
• Salesforce holds the account Heroku is a Salesforce, Inc. product run from San Francisco, so the CLOUD Act reaches the provider regardless of which region the dynos run in. This is the ordinary argument and it is true here, but it is not the interesting one. What is interesting is that the platform was acquired in 2010 and the pricing, the free tier and the runtime generations have all since been decided by a company whose main business is something else entirely.
• The 2022 incident is still the best argument for reading the postmortem On 7 April 2022 a threat actor obtained access to a Heroku database and downloaded stored customer GitHub integration OAuth tokens, then exfiltrated a database of usernames and hashed and salted passwords. Heroku revoked every GitHub token on 15 April and forced a password reset on 5 May. The handling was reasonable; the lesson is that a platform which holds your deployment credentials is a single point of failure for every repository it can reach.

What you have to replace, not just match

Heroku sells three things in one bill, and which of them you are actually buying decides whether this move is an afternoon or a quarter.

The first is git push deployment: a buildpack detects the language, builds a slug and runs it, with no Dockerfile and no pipeline to write. The second is the managed data services, Postgres and the key-value store, attached to the application with a config var rather than provisioned separately. The third is the add-on marketplace, where logging, monitoring, mail and scheduling arrive as one-line installs.

Clever Cloud answers the first two directly and is the only European product that really does. Northflank answers a different question, which is what you do when the application has outgrown a slug and wants containers, GPUs and its own cloud account. Nobody here answers the third, so make the list of your add-ons before you decide, because that list is the actual migration.

The alternatives compared

How each alternative compares to Heroku

What actually breaks when you switch

The build is where migrations fail, not the runtime. Heroku buildpacks quietly handle asset compilation, native extensions and language version pinning that nobody on the team ever wrote down, and those decisions surface as a failing build on the first day somewhere else. Do a throwaway deploy of the current main branch before you plan anything, because that build log is the real scope document.

Then comes the add-on tail. Papertrail, Scheduler, a mail provider, an error tracker, a Redis-backed queue: each is a separate account, separate credentials and a separate invoice once it leaves the Heroku bill, and at least one of them will turn out to be doing something load-bearing that nobody remembers configuring. Enumerate them from the dashboard rather than from memory.

And plan the database cutover properly, because it is the only irreversible step. A Heroku Postgres follower can be promoted to keep downtime short, but the follower has to be created before you start and the application has to tolerate a read-only window. Teams that skip this end up doing a dump and restore at two in the morning with the site down, which works and is a bad way to spend a Tuesday.

Is Heroku actually being wound down?

No, and repeating that is the fastest way to lose an internal argument to somebody who has read the changelog. Fir went generally available, Heroku recommends it for new and actively developed applications, and the Dev Center documentation is being updated through 2026.

What is true is that the platform has been reshaped around a different customer. The features arriving first on Fir are Cloud Native Buildpacks, OpenTelemetry observability and Graviton support, which are the concerns of an engineering organisation rather than of one developer with a Rails app. The features still missing from Fir are the ones that made Heroku cheap.

So the accurate version of the complaint is not neglect. It is that the product is being aimed above you, and the price follows the aim.

What does leaving actually cost in engineering time?

Less than most platform migrations, because Heroku taught everyone the twelve-factor habits that make an application portable. Configuration is already in environment variables, logs already go to stdout, the process model is already declared in a Procfile, and the database is already reached through a connection string. That is most of the work done years ago.

The cost lands in three places. The buildpack does things you never wrote down, particularly around asset compilation and native dependencies, and those surface as build failures on the new platform. Anything scheduled through Heroku Scheduler needs recreating. And every add-on is a separate small migration with its own credentials.

A plain web application with a Postgres database moves in a day or two. An application with six add-ons, a Redis-backed job queue and a review-apps pipeline takes a fortnight, most of it spent on the pipeline rather than the app.

How do the numbers compare once you are honest about both sides?

For a small application Heroku is genuinely cheap and it is worth saying so. A Basic dyno at $7 with a Postgres Essential plan between $5 and $20 runs a real side project for under $30 a month, and Clever Cloud from around EUR 1.49 undercuts it while Northflank's Sandbox tier gives two always-on services free.

The crossover is at the point where you need the things Heroku sells above the entry tiers.

Standard-1X is $25 a dyno, Performance-M is $250, Private-S is $125 and Shield-S is $150, and Postgres Standard plans start at $50 and run to five figures. That is where usage-based European pricing starts winning by a lot rather than a little: Northflank bills CPU at about $0.01667 per vCPU-hour and memory at $0.00833 per GB-hour, pro-rated to the second, with no per-seat fee.

The number that actually decides it is egress, and neither side advertises it. Check what you move out per month before you compare anything else.

Do I have to leave the buildpack model behind?

Not if you go to Clever Cloud. You push a Git repository, it detects the runtime from among fifteen or more languages, builds it and runs it, and a managed PostgreSQL, MySQL, MongoDB or Redis add-on attaches to the application rather than being provisioned as a separate thing with a connection string to copy. That is the Heroku experience, in French data centres, from a French company.

Northflank takes you the other way deliberately. It builds containers, runs them on Kubernetes, and can do it inside your own AWS, GCP or Azure account, or on Civo, Oracle or CoreWeave. If your application already has a Dockerfile, that is a promotion. If it does not, you are writing one, and that is the real cost of choosing it.

The useful question is whether the buildpack was the reason you were on Heroku or just the thing you got used to. For a small team with one application it is usually the reason. For a team with five services and a CI pipeline it usually stopped being the reason some time ago.

Which one to pick

If what you want back is the Heroku of 2015, Clever Cloud is the answer and it is the only real one in Europe. Git push, an attached database, French data centres, and a company that will put in writing that extraterritorial law does not reach your data. For most teams leaving Heroku for jurisdictional reasons, the search ends here.

If the application has already outgrown the slug, Northflank is the better fit and will feel like a step up rather than sideways. Containers, real pipelines, GPUs, and the option to run all of it inside your own cloud account, which Heroku has never sold at any price.

If you are leaving because of the missing free tier, be realistic: Northflank's Sandbox is the closest thing to what you lost in November 2022, and Clever Cloud at around EUR 1.49 a month is cheap enough that the difference is not worth a weekend.

And if the reason you are reading this is that you heard Heroku is being shut down, it is not. Fir is generally available and recommended for new applications. Leave because the pricing and the shape no longer suit you, which is a defensible reason, rather than because of a rumour a colleague will check.