Honest answers.

Deterministic anti-cheat for expensive actions: one embedded script reads the shape of interaction — cadence, integrity, coherence — and your backend gets a confidence-qualified verdict — allow, limit, step up, or deny — before the action runs. No puzzle, no LLM.

Ever. There is nothing for a visitor to see, click or solve — which is also friendlier for the people traditional captchas lock out. If the evidence is thin, the verdict says so, and your policy decides whether to allow, limit, step up, or deny.

No. The pipeline is deterministic: the same evidence always produces the same verdict, with stable reason codes you can log and audit. No model drift, no black box, no inference cost — not an AI guessing about an AI.

Counts, never contents. We derive rhythm from typing (never keys), shape from pointer movement (never coordinates), and cadence from scrolling. We never read page content, prompts, or form values, keep no cross-site profile, and drop the raw signal after deriving the verdict. We keep the proof you are human — a realness signal, not a record of what you did — and never sell it.

Yes — a fully coherent cold bot can pass, and any vendor who claims otherwise is selling you a guarantee that cannot exist. That is why the verdict is "likely human," confidence-qualified, and why it pairs with your policy and outcome feedback instead of pretending to be automation-proof identity.

Planned Launch is $499/month for one app and 100,000 unique successful evaluated actions; Growth and Enterprise scale from there. Billing is per unique successful evaluated action — collection, simulation, retries and failed requests are never billable. Self-serve checkout is not live yet; plans start through us.

One script tag on the page, one GET from your backend. Most teams run in shadow mode the same afternoon and turn on enforcement once they have seen a week of recommendations against real traffic.

Your product keeps working — the embed is passive and the verdict read is a normal API call you can time out. We publish live status and recommend failing open (or to your own rate limits) rather than blocking humans on our availability.