Try it with your own data
Upload a CSV, tell Kirelta which rows were healthy, and get a real verdict from the real engine — before you write a single line of integration code.
Your file is read in your browser. Parsing, column detection and the split all happen on your device, and the file is never uploaded. When you press Run, only the numbers in the columns you selected are sent to the API. Kirelta keeps the model it fits from them, never your file; the Trust & Security page lists exactly what a model holds.
Which rows are the baseline
Baseline — rows Kirelta learns as healthy
—
The earliest rows in the file. This is what "normal" will mean.
To check — rows it will judge
—
The rest. Kirelta has never seen these.
Drag to move the split. If your file is ordered by time, this is exactly the real question: does last month still look like the months before it?
Drag to move the split. Heads up: “§COL§” repeats in blocks rather than being scattered, so this file may be sorted or grouped by it. If so, the split compares categories, not earlier against later. Shuffle the rows first if that is not what you want.Your API key
Held in this tab only, never stored. Get a free key — 2 models and 1,000,000 predictions a month on the free plan, no card.
What the engine said
This ran against the real engine — the same one your production calls would hit. Kirelta kept the model it fitted as demo, and your file stayed on your device. From code it is two calls: see the quick-start.