Free
$0
forever · no card
Try it and learn it. Enough to wire Kirelta into a real service.
- 2 models
- 1M predictions / month
- 1 user
- 1 API key
- Community support
- No CSV export of the alert log
No card. No trial clock.
Pricing
The free plan is not a trial — it doesn’t expire. Upgrade when you have more models, more teammates, or more predictions than it holds.
$0
forever · no card
Try it and learn it. Enough to wire Kirelta into a real service.
No card. No trial clock.
$99/mo
$82.50/mo
billed monthly
For small and mid-size teams running Kirelta in front of production traffic.
$349/mo
$290.83/mo
billed monthly
For companies running several projects or teams on Kirelta.
Talk
annual · invoiced
Run it inside your own perimeter, or at a volume the plans above don’t cover.
Pricing depends on scope.
Plans differ in scale and support, not in what Kirelta does. Every plan, the free one included, gets the whole product.
No. It doesn’t expire and we don’t ask for a card. If two models and a million predictions a month is all you need, stay there.
One row assessed by /assess. Sending 500 rows in one call counts as 500 predictions; sending them one at a time is still 500. Billing is based on predictions processed per calendar month, not the number of API calls or the size of any one request.
Your account is never suspended and you’re never locked out of the dashboard. New calls to /assess return a 429 with the message the API actually sends, in English: You have reached your monthly prediction limit. Upgrade your plan to continue processing predictions.
Everything you’ve already stored — models, history, keys — stays exactly as it was. The counter resets at the start of the next calendar month, or upgrade any time to keep going immediately.
The workspace owner invites teammates by email from Account & keys. Everyone on a team shares the same models, API keys and billing — seats are logins, not separate accounts with separate limits.
Kirelta receives the numbers you send to /fit and /assess — it can’t compute anything without them. It does not receive your model, your code or your labels. What it keeps from a fit is the model it builds: for most models, summary statistics and, for each column on its own, the values set aside to calibrate it, sorted. If you upload a baseline as a CSV, the file itself is kept, up to your plan’s limit, so it can be rebuilt. The privacy policy lists everything kept, and for how long.
No — and this is deliberate. Kirelta produces a verdict, the evidence for it, and a recommendation. Acting on that is your call, in your code. There is no webhook that reaches into your infrastructure, and we will not add one quietly.