Pricing
You pay for the stepsthat think.
Deterministic steps are free to run — all of them, on every plan, with no monthly allowance to watch. What you pay for is the number of workflows you operate and the number of steps that call a model. Most workflows call one on a good day.
€0
Free
For a first workflow, and for the ones that genuinely never need a model.
- 3 workflows
- Unlimited deterministic steps
- 200 AI steps a month
- Schedule and event triggers
- All castomAPI reference data
- Simulation, version pinning, run logs
€24 / month
Pro
For a business running its own operations on this.
- Unlimited workflows
- Unlimited deterministic steps
- 3 000 AI steps a month
- Webhook triggers
- Priority support
- Everything in Free
€79 / month
Team
For agencies and for teams where more than one person edits.
- Everything in Pro
- 15 000 AI steps a month
- Shared workflow library
- Role-based access
- Unlimited workflows
- Priority support
Prices exclude VAT. Billing is monthly and you can change plan or stop at any time; there is no minimum term and no notice period. One castomo account and one wallet cover every product in the family, so if you already pay for one of the others, this joins the same invoice.
Why the meter is on AI steps.
Because that is where the cost actually is. A deterministic step — fetch, transform, branch, wait, write, send — costs us a fraction of a millisecond of a machine we already run. Metering it would cost more to administer than it earns, and it would make you hesitate before adding a step that validates an address, which is precisely the wrong instinct to encourage.
A step that calls a model costs real money on every invocation. So that is the meter, and the effect is a pricing model that quietly pushes you towards the better design: use a rule where a rule is enough, and spend a model call on the two steps that genuinely need judgment.
The industry is currently moving the other way, putting a model on every step because that is what is being sold. It makes for an expensive reconciliation.
How far an allowance goes.
An AI step is one invocation of a model at one point in one run. A workflow with a single classification step, running two hundred times a month, uses two hundred AI steps — the rest of that workflow, however long it is, costs nothing to run.
- 1Order workflow640 runs × 7 steps. One AI step each — the customer note.
- 2Basket workflow310 runs × 6 steps. One AI step each.
- 3Nightly reconciliation30 runs × 5 steps. Zero AI steps. Entirely deterministic.
- 4Supplier feed30 runs × 4 000 rows. Zero AI steps.
- ΣTotal950 AI steps of 3 000. Roughly 130 000 deterministic steps, all free.
€24 for the monthAn illustration, not a promise about your shop
Questions people ask before signing up
The awkward ones, answered
What happens when I run out of AI steps?
Workflows keep running. Deterministic steps are unaffected, and a run that reaches an AI step stops there with a clear reason rather than silently skipping it and producing something wrong. You can add steps or change plan, and the stopped runs resume.
Do simulations cost anything?
An AI step invoked during a simulation is a real model call, so it counts. Everything else in a simulation is free, as it is in a real run.
Is there a per-run or per-task charge?
No. That is the usual model in this category and it is the reason people build fewer, larger, less readable workflows. Here a run costs nothing and a step costs nothing unless it calls a model.
What does a payment step cost?
Taking a payment through the family's payment service is 2% plus €0.25 per transaction, and the money goes to your own account. That is a charge on the transaction, not on the workflow step.
Is reference data really free?
Yes, on every plan including Free. Sixty-five endpoints, no authentication, no quota, no per-lookup charge. It is public reference data served by a platform we already run for the rest of the family.
Can I bring my own model provider's key?
Yes. Steps that run against your own key are billed by that provider rather than counted against your allowance. Most people do not bother, but if you have negotiated rates or a compliance requirement about where inference happens, it is there.
Do you have an annual discount?
Ask. It exists, and it is not a headline number on this page because a discount that requires a year of commitment should be a conversation rather than a checkbox.
What about a plan larger than Team?
Write to us with the volume you are expecting. There is no self-service tier above Team because above Team the interesting questions are about data residency and support terms, not about a number of steps.
What is not on this page
A discount that expires this Friday. A tier whose price is hidden behind a form. A per-seat charge that turns showing a colleague what you built into a purchasing decision. If any of those appear here later, it will be because something changed in the business, and it will be written down rather than discovered at renewal.
The genuinely honest bit: the Free plan is not a trial. Three workflows and two hundred AI steps a month is a real amount of work, and a small business that automates its three most annoying tasks and never pays us anything is a perfectly good outcome. The plans above exist for when three workflows stops being enough — which, if this works for you, it will.