What Kirelta actually does
No item on this page is aspirational — everything here is running code you can test yourself at try.html right now.
The engine
Anytime-valid drift detection
Evidence accumulates across the rows of a batch rather than being re-decided from a single snapshot, and the alarm threshold carries a real, provable false-alarm bound over the whole run — not a tuned-until-it-looked-right one. The measured rates are published on the calibration page.
Three verdicts, precisely defined
UNTRUSTED
the drift alarm fired, or a large share of the batch was flagged.
DEGRADED
more of the batch is falling outside the trained range than a healthy stream should produce, but no alarm yet — the state most tools miss entirely.
TRUSTED
neither of the above.
Feature attribution
Every UNTRUSTED or DEGRADED verdict comes with which specific inputs moved and by how much — not just "something changed."
Independent evidence layers
Input drift, output stability (when you send your model's own predictions), and baseline-quality suspicion are tracked separately and shown separately — a healthy input layer doesn't hide a suspect baseline.
The decision workflow
Decision Journal
Every verdict change writes a permanent, evidence-backed entry — not just a log line. Review, approve, reject, or archive, with a note that's recorded against your name.
Tamper-evident evidence
Each decision's evidence snapshot is sealed with a content hash at write time and re-verified on every read. If the stored record had been altered, the page says so.
History & trends
Per-model trust-over-time charts, retained for your plan's window (7–3650 days depending on plan — see pricing).
Real audit log
Key creation, revocation, team changes, sign-ins, and account deletion are all recorded per tenant, append-only.
For developers
Three calls, no framework required
/fit once, /assess per batch, branch on the verdict. Full reference and copy-paste examples in the docs.
Official Python client
Zero dependencies beyond the standard library. pip install kirelta.
Webhooks
Get notified the moment a verdict changes, instead of polling.
Live console
Paste an API key and watch a model's real status update — no account required to look.
For teams
Real team seats
Invite teammates by email; seat limits scale with plan.
GDPR-complete deletion
Account deletion removes the account, its models, its keys, and its decision history — verified by an automated test after every deletion, not a manual checklist.