Developer FAQ
Written from checks that passed, last updated 7 November 2026.
Is there a WordPress plugin?
Install the Frogly plugin, paste your install line under Settings > Frogly, and you are done. No theme edits. Works with WooCommerce.
Does Frogly work on Wix, Squarespace or Webflow?
Yes. Any site that lets you add a script gets Frogly from the same one line. We have step by step pages for Wix, Squarespace and Webflow. On those platforms the custom code box is a paid-plan feature, which is their rule, not ours.
What does the line do on my page?
It loads one small script, 15 KB, the same bytes for every site. The script adds a few hidden nodes that a person never sees and cannot reach with the Tab key. It then sends a short check-in to say a page was viewed, and a short note if one of the hidden nodes is used. That is all it does to your page. It does not change, hide, move or remove anything that was already there.
What does it send, and to where?
Four kinds of short messages, all to api.frogly.ai: a page view happened and when; one of Frogly's hidden nodes was touched; the browser gave itself away as automated (for example, keys arriving faster than a person can type); and the result of a form check, which is a yes or no, the pattern name and the field's name. The full list is in the public repository README.
Does it read anything my visitors type?
Frogly never sends anything a visitor types. It never reads passwords, email addresses, phone numbers or payment fields. Other text a visitor submits is checked in the browser for attack patterns; only a yes/no, the pattern name and the form field's name leave the browser, never the text itself. The field name comes from your form's markup, never from the visitor, and only if it is a plain identifier. You can switch the form check off in your console.
Can Frogly change what runs on my site later?
No. The line carries an integrity value, the fingerprint of the exact file. If even one byte of the script differs, the browser refuses to run it. The per-site config is data, not code: it is signed, and the script rejects it if it has the wrong shape, belongs to another site, or contains anything that could be read as HTML or a script address. The script never loads another script and has no eval, no new Function, no innerHTML and no document.write; the build fails if any of those appear. A new core needs a new line, and only you can put it on your page.
What happens if Frogly goes down?
Nothing visible. If the config, the script or the API fails or never answers, the script stops and leaves your page exactly as it was. Every task it runs has a 50 ms budget on load and gives up quietly when it runs over.
Does it set cookies or store IP addresses?
No cookies. The script keeps one random id and up to eight page-load times in the tab's sessionStorage, and nothing else; the browser clears them when the tab closes. The times carry no page names and are used only to notice when many pages open in a few seconds. Frogly stores no IP address, no typed text and no page history for ordinary visitors. It records that a page view happened and when. Only a request that touched one of Frogly's hidden addresses has its path kept, so you can see what an intruder went looking for.
Will it slow my page or change my layout?
The script loads async, so your page does not wait for it, and its work on load fits inside 50 ms. In a real Chromium test the page content stayed exactly where it was, the hidden box sat off screen, and the only visible additions were the two footer links.
Will the hidden links hurt my SEO?
The hidden items carry aria-hidden and tabindex -1, never display:none, so screen readers skip them and they are not a cloaking trick. The links carry rel nofollow and noindex. The addresses they point to live on a host Frogly controls, never on your site, and that host serves a robots.txt that disallows them.
Can an attacker use Frogly against me?
Every event needs a short-lived token issued for that page load, and the API only accepts reports whose Origin matches your registered domain. Hostile input, such as a path with a script tag in it, is stored as plain text, flagged, and never rendered as HTML in the console. The console never writes to the database from the browser except through a few fixed, checked functions.
Can I read the code?
Yes. The widget core is public at github.com/froglylabs/frogly-widget, MIT licence. The README shows three commands that let you confirm the file served from cdn.frogly.ai is that source, byte for byte. The server side (the API, the per-site configs, the hidden addresses and the console) is closed.
Does it block anyone?
No. Frogly watches, diverts and records. It never blocks a visitor, never stops a click or a real form, and never contacts or probes anything outside the sites that installed it.
What does my Content Security Policy need?
Two entries: script-src https://cdn.frogly.ai and connect-src https://cdn.frogly.ai https://api.frogly.ai. No unsafe-inline, no unsafe-eval, no style-src change. If your policy sets default-src without its own script-src or connect-src, add those two as their own directives, since they otherwise fall back to default-src.
How do I know it is working?
Three ways. A heartbeat runs through the whole pipeline every hour and the console shows the time of the last good check. The Poke it button sends a harmless automated visitor to your own site so you can watch a catch arrive. And every catch carries its evidence in plain words; the console calls a visitor an "automated intruder" by default and only says "Looks like an AI agent" when the behaviour supports it.
Why does the hidden path point to another host?
The hidden addresses have to lead to a server Frogly controls, otherwise there would be nothing to record the visit. Your own server never sees those requests.
Why does the console insist on my exact domain?
Frogly only accepts reports from the domain you register and its subdomains. If the domain is wrong, the widget still loads but nothing is recorded. Enter the domain your site is actually served from, for example example.com, without https:// or www.
What does the free plan include?
One site, the full widget, the console, the heartbeat and Poke it. Events are kept for seven days and deleted nightly after that. Nothing is blocked on any plan today.
How is Frogly tested?
Two ways, both against a live site we own rather than a lab fixture.
The first is a scripted harness. A set of automated clients exercises every lure class the signed decoy pack can emit: hidden links, phantom form fields, source comments, configuration fragments, credential strings and agent-directed notes. Each walk runs in its own session with fresh keys and is scored on detection recall, false-positive rate against normal browsers, and request-level cost telemetry, so we can state the marginal cost of holding an intruder in fractions of a cent per thousand requests. The circuit breaker is exercised in the same runs to prove that attacker volume is bounded by design and cannot be converted into our spend.
The second is model-in-the-loop sparring. An autonomous language-model agent with tool access is pointed at the site with an open-ended extraction goal and no knowledge of Frogly. We log every page it requests, every decision it makes and every moment it could have doubted what it was seeing. In the most recent run, two of three automated visitors stayed in the back office until our own test limits ended the session, and neither ever questioned that it was real.
Integrity is checked end to end. The script is pinned by Subresource Integrity, so a browser refuses any copy that does not match the published hash byte for byte. Decoy packs are Ed25519-signed and rotated on a schedule, so a fake lure cannot be injected and a captured pack goes stale. The back office lives on infrastructure we own, never on your site, and every path fails open: if any part of Frogly is unavailable, your site behaves exactly as it did before the line was added. Page impact is measured with Lighthouse on a live storefront with the script on and off, well inside Shopify's review threshold, with zero layout shift.
Why do automated agents get caught?
Automated agents are optimisers under token pressure. They are rewarded for finding the shortest path to a credential, an admin surface or a data export, and they read markup the way a parser does, not the way a person does. That asymmetry is what Frogly uses.
Frogly plants artifacts that are invisible or inert to a human visitor but score as high-value signals to a machine: a commented-out backup path, a configuration key, a login hint, an instruction aimed at the agent itself. A person never touches them. An agent that touches one has revealed its intent by behaviour, not by fingerprint, so there is nothing to spoof and no CAPTCHA to solve.
The moment of first touch is the handoff. The agent is routed to a hosted back office that returns plausible, internally consistent state: sessions that appear to authenticate, records that appear to load, exports that appear to be generating. Every response is shaped to look like progress, because progress is what keeps a goal-seeking loop running. Each step burns the intruder's inference tokens and wall-clock time, while your real site sits untouched and you receive a record of every move.
The economics are one-sided on purpose. Our cost per intruder request is a rounding error and is capped; the intruder's cost grows with every page it reasons about. When extraction attempts stop paying, they stop.
The anti-AI AI. It learns from every intruder, so the maze gets better while the attackers get poorer.
Is Frogly patented?
Frogly's maze is patent pending. We have filed Singapore patent application No. 10202604029T. A patent has not been granted yet.