How it works
A workflow is a sequence of steps.
That is the whole model, and it is deliberately unexciting. Something starts a run. The run walks a list of steps. Some of those steps think, most of them do not, and a few of them wait. When the list ends, so does the run — and every decision it took is written down.
Most automation tools ask you to draw that sequence before you know what it is. You are handed a blank canvas and a list of four hundred integrations, and the first twenty minutes go on remembering what the thing you wanted was actually called.
castomFLOW inverts it. You write the sentence; the workflow arrives built. Everything after that is editing something concrete, which is a very different job from inventing something abstract.
The three kinds of step.
Deterministic steps fetch, transform, branch, wait, write and send. The same input produces the same output, every time. They are fast, they cost nothing to run, and they are dull in the way infrastructure is supposed to be dull.
AI steps classify, draft, summarise and choose. They take structured input and return structured output, so a model is never left to freelance beyond the shape you defined. These are the steps you pay for, and the pricing exists to make you thrifty with them.
Flow steps do not act on the outside world at all. They wait, branch on a condition, split traffic between two paths, hold a run until a quiet-hours window closes, or mark a goal as reached and stop.
A single workflow moves freely between all three. The task decides which parts think.
- ⚡TriggerWebhook from the shop:
order.paid, 289,00 zł. - 2Validate addressDeterministic. Postal code matched against reference data.
- 3Raise invoiceDeterministic. Fields came from the record; nobody typed them.
- 4Write the noticeAI step. Two sentences, in the customer's language, in your tone.
- ?ConditionIs it after 21:00 local time? No. Quiet hours not in play.
- 5SendE-mail out, delivery state tracked back onto the run.
Elapsed: 4.1 sPaid steps: 1
Every process has a shape.
Nobody has ever drawn yours.
The engine, part by part
Six pages.
Then you know it all.
There is no hidden complexity below this. Read these six and you have read the product.
Describe it
What actually happens in the two minutes between your sentence and a built workflow — and what you get back.
Triggers & actions
Every way a run can start, everything a step can do, and how waits, conditions, splits and goals shape the path.
The castomAPI advantage
The steps that exist here and nowhere else, because the platform behind them belongs to the same family.
Connecting anything
HTTP in, HTTP out, authentication, pagination, rate limits, signatures — and the system that only speaks CSV.
Runs you can trust
Simulation, version pinning, retries that cannot double-send, and a log written for a human reader.
What it costs
Deterministic steps are free to run. You pay for the steps that call a model and the number of workflows you operate.
What this is not.
It is not a general-purpose programming environment. If a task needs a real language, write it in a real language and call it from a step over HTTP — that is a supported and frequently sensible answer, not a defeat.
It is not an agent that decides for itself what to do next. An AI step is invoked at a point you chose, given the input you chose, and constrained to a shape you defined. The sequence belongs to you. That is a deliberate limit, and it is the reason a run is reproducible.
The honest bit: a workflow that has been built for you in two minutes is a good draft, not a finished product. It will have a wait you disagree with and wording you would not have used. Twenty minutes of editing is normal, and it is twenty minutes rather than an afternoon because the shape is already there.