Workflow automation
Self-hosted Zapier alternative: what each one costs to run
Every comparison you can find is written by one of the tools. They agree on the feature grid and go quiet on the part that decides the bill, which is who keeps the server alive and what happens the night it stops.
The search that brings people here is usually triggered by an invoice. Per-task pricing charges for volume rather than for value, so the bill grows with a business that is working, and at some point somebody asks whether a server would be cheaper. Sometimes the trigger is a data question instead: a procurement or compliance review wants to know which third party holds customer records in transit, and the honest answer about a hosted automation platform is not one anybody enjoys writing down.
Both are good reasons to move. Neither survives a plan that assumes self-hosting is free.
The options, and what each is actually built for
These tools get listed together because they all draw boxes and connect them. They were built for different jobs, and picking on the feature grid is how teams end up rebuilding twice.
| Tool | Licence family | Built for | Where it hurts |
|---|---|---|---|
| n8n | Fair-code (Sustainable Use License), free to self-host for internal use, not OSI open source | Broad SaaS integration with an escape hatch into code, so a workflow can drop into JavaScript when the node runs out | The licence forbids reselling it as a hosted service. Read the terms if you are an agency planning to run it for clients |
| Activepieces | Conventional open source core, with paid tiers around it | Teams where non-engineers build and own the workflows, because the builder is the cleanest of the group | A smaller connector library, so the odds of writing your own HTTP calls are higher |
| Automatisch | AGPL, conventional open source | Cases where the licence question has to have a boring answer | The smallest connector set here. Fine for a handful of mainstream apps, thin beyond that |
| Windmill | AGPL core with a paid enterprise edition | Engineering teams who would rather write Python or TypeScript and get a UI and a scheduler around it | It is not a Zapier clone. Non-engineers will not maintain what gets built in it |
| Node-RED | Apache 2.0 | Event and device flows, which is where it came from | SaaS integrations are community-contributed and uneven, so business automation is swimming upstream |
Licence terms on this list have changed before and will change again. Check the current text on the project's own repository before you commit an estate to it, rather than trusting a comparison table, including this one.
The running-cost line items nobody puts in the comparison
The arithmetic that makes self-hosting look free compares a monthly platform invoice against a monthly server invoice. Those are not the same list of items. The platform fee was buying things that do not disappear when you stop paying it, and here is the full list so you can price your own version.
| Line item | What the platform was doing | What you own after the move |
|---|---|---|
| Compute | Included | A VPS or container, sized for peak concurrency rather than average load |
| Database | Included and invisible | Postgres, with growth from execution history that has to be pruned on a schedule |
| Backups | Included | Your own, and only real once you have restored one into a scratch environment and watched it work |
| Upgrades | Applied for you | A version bump you schedule, test and roll back yourself when a node changes behaviour |
| TLS and domains | Included | Certificates and renewal, usually automated, occasionally the thing that breaks at renewal |
| Monitoring | An email when a run failed | An error workflow, an alert channel and somebody who reads it |
| On-call | Their problem | Yours, at three in the morning, when the queue stops moving |
Add those and self-hosting still wins for most teams past a few thousand tasks a month. It stops being free. The fair framing for a finance conversation is that you are trading a variable per-task fee for a fixed infrastructure cost plus a few hours of engineering attention a month. If you want the band rather than the argument, the automation cost calculator runs it off the published 2026 price list.
The failure mode that costs more than the licence
A hosted platform tells you when a run fails. It sends an email, holds the failure and lets you replay it. A self-hosted instance with no error workflow attached fails into the execution log, where nobody is looking. That is the single most common way a migrated estate goes quiet without anyone noticing, and the damage is measured in the weeks before somebody spots it rather than in the minutes of downtime.
We have watched this from the inside. A weekly reporting workflow we run for a US email-marketing agency client failed quietly rather than loudly, and the fix was moving it somewhere that shouts. Wire the alerting before you migrate the first workflow, not after the last one.
What we run ourselves
Digiton runs its own operations on owned infrastructure with scheduled jobs rather than metered third-party automation, so this is a bill we pay in our own time instead of a position we sell. Two things came out of that worth passing on.
A visual canvas earns its place when a non-engineer has to read the logic, and stops earning it when the logic is a script wearing thirty nodes. We moved a large share of our own recurring jobs to plain scheduled scripts for that reason. And the operational risk that actually bites is silence, not downtime. The mechanics of why a self-hosted instance gets slower as you add capacity are in why self-hosted n8n slows down as you add workers.
A short checklist before you commit
- Count the distinct services you need re-authenticated, not the number of workflows. That number sets your calendar, because each one needs an admin on the other platform.
- Check every service against the tool's connector list. Anything missing is HTTP calls, pagination and token refresh you write and maintain.
- Read the licence if you plan to run it for clients rather than yourself.
- Decide who is on call before you decide which tool. If the answer is nobody, stay on the hosted platform and negotiate the invoice instead.
- Stand the instance up with error handling and alerting wired first, then migrate.
If the destination is n8n specifically, the hour-by-hour breakdown by workflow type is in what a Zapier to n8n migration actually takes. If the driver is data protection rather than cost, the retrieval side of that argument is in GDPR-compliant RAG.
Digiton is an AI consulting firm in Lisbon. We build and operate automation for clients across 8 countries, and we will tell you when the migration is not worth doing.
Frequently asked questions
What is the best self-hosted Zapier alternative?
For most teams replacing a Zapier estate, n8n, because it has the widest connector library of the self-hostable options and the largest community, so a broken node usually already has a documented answer. Activepieces is better when non-engineers build the workflows. Automatisch fits when the licence has to be conventional open source. Windmill fits engineering teams who would rather write code.
Is n8n really open source?
No, not in the OSI sense. n8n uses a fair-code Sustainable Use License. You can self-host it free for internal business use, and you cannot resell it as a hosted service. If you are an agency planning to run instances for clients, read the current licence text before you build a business on it. Automatisch and Node-RED use conventional open source licences.
Is self-hosting actually cheaper than Zapier?
For most teams past a few thousand tasks a month, yes, but only once the comparison is fair. The platform fee was covering compute, the database, backups, upgrades, TLS, monitoring and an on-call answer. Price all seven and you are trading a variable per-task fee for a fixed infrastructure cost plus a few hours of engineering attention a month.
What breaks most often on a self-hosted automation server?
Silent failure. A hosted platform emails you when a run fails and holds it for replay. A self-hosted instance with no error workflow attached fails into the execution log where nobody is looking, so the estate goes quiet for weeks before anyone notices. Wire alerting before you migrate the first workflow.
Can I import my Zapier workflows automatically?
No. None of these tools read Zapier exports, and Zapier does not publish a format designed to be read by a competitor. Every workflow gets rebuilt by hand using the old one as a reference, which is why migrations are budgeted per workflow rather than per account.
How much server do I need to run one?
Size for peak concurrency rather than average load, because the failure is a queue that stops moving during your busiest hour rather than a monthly average that looks fine. Execution history in the database grows quietly and needs pruning on a schedule, which is the maintenance item most teams find out about the hard way.
When should we stay on Zapier?
When total automation spend is under a few hundred a month and nobody on the team wants to own a server, or when the cost is concentrated in one expensive workflow that could simply be rewritten. Data residency is a separate argument and is often sufficient on its own even when the cost case is not.
Related
Want the number before the call?
The cost calculator gives you a build band, a monthly running cost and a payback range from our published 2026 price list. No email needed to see it.
Open the AI automation cost calculatorTell us the operation, get a scoped answer
One line about the work you would hand over. We reply within one working day with the band it falls in and whether it is worth building at your volume.