47 features
Behavioural and network signals feed a gradient-boosted model with Platt calibration, so 0.80 means about what it says.
Features
Scoring, adaptive work, and verification you control, without sending your visitors through puzzles.
Scoring
Behavioural and network signals feed a gradient-boosted model with Platt calibration, so 0.80 means about what it says.
A second layer of rules can reduce a score but never raise it. Webdriver, headless user agents, datacenter ASNs and Tor hit a hard ceiling.
If the model artifact is missing or corrupt the engine falls back to a transparent linear model instead of erroring.
Proof of work
Difficulty scales between 16 and 24 bits with the risk of the request, so honest visitors barely notice it.
The work runs in a Web Worker, so the page stays responsive while it solves.
The solution is checked on the server and a solution is accepted once.
Products
Invisible. The widget scores in the background and you decide what to do with the number.
Adds a checkbox. A verified proof of work plus a real click clears the test, and a scripted click fails the score.
Run a site in test mode while integrating, or shadow it to watch verdicts before enforcing them.
Control
Generate a secret when you need one, revoke it when you do not. Secrets belong to the site.
Bind a site to its domains and set the pass threshold and difficulty that suit it.
Hashed logs with 7 to 90 days of retention, usage alerts, and webhook delivery with retries.
Start on the free tier. Create a site, copy the sitekey, ship.