How it works

Agent recordings

Visits by automated browsers are recorded as a masked replay of your page. People's are not, with one edge case.

A visit by an automated browser is recorded as a replay of your page: its structure and the changes to it, as the agent saw them, with what was typed and nearly all text masked. You watch it from the visit in the dashboard ("Watch the real page"), to see exactly where an agent got stuck.

Agents are recorded, people never

The recorder is a separate piece of code. The browser part loads it only when the page's own signals say the browser is automated: automation flags or the traces automation tools leave, a headless browser, or software graphics rendering. Weaker signals (no graphics information, a browser without a consumer brand, a denied notification permission) never load it, even together. A person's browser does not download it, with the edge case below.

The gateway also refuses recording for any visit it has judged to be a person.

What a recording contains

The recorder captures the page's DOM and its changes (with rrweb), and masks as it captures, in the browser, before anything is sent:

  • Every input a visitor can type into is masked, always: a typed card number, email or address never leaves the page.
  • Addresses are rewritten. Hidden form fields and link, image and form addresses are rewritten by the address rules, and the page address in a recording keeps only its path and campaign parameters (on arrival since October 16, 2026; in the recorder from 1.0.4).
  • Every text node is masked, except the label of a real control (a button, or a link with little inside it), which stays readable so you can see what the agent clicked. Emails, phone numbers and runs of four or more digits are masked inside those labels too.
  • Iframes are left out, so embedded payment fields and other frames are never captured.
  • Mouse movement is sampled, not recorded in full.

Recordings never include cookies, stored data or network requests.

Your controls

Three attributes change what the recorder captures inside an element:

AttributeEffect
data-plx-unmaskText inside it is recorded as it is. Use it for content that is safe to show, such as product details, so the replay reads
data-plx-maskMasks everything inside, control labels included
data-plx-blockLeaves the element out of the recording entirely
HTML
<section data-plx-unmask>…product description, price, delivery…</section>
<div data-plx-block>…an account area…</div>

Turning recording off

With the npm package:

TypeScript
init({ publicKey: "pk_…", recordAgents: false });

The recorder is then never loaded, and nothing is recorded. The pinned tag has no such switch: it reads only its key and its consent setting, so with the tag, agent visits are always recorded.

Withdrawing consent (consent(false)) stops a running recording at once, and nothing buffered is sent.

Bounds

  • The recorder stops after 30 minutes, 40 chunks or 5 MB of recorded data per visit. The gateway enforces the same bounds itself, and accepts no chunk for a visit idle for more than 35 minutes.
  • Chunks are compressed in the browser and sent to https://gateway.parlox.io/v1/r; the last one, sent as the page is left, goes uncompressed.
  • Recordings are deleted automatically after a limited time, and never kept longer than the site's visit data (Where data lives and how long).
  • A recording is never drawn on the dashboard's own address: it plays in a separate player page, with a short-lived ticket for that one visit.