How Parlox works
AI agents reach a store in two ways, and each is seen by a different part of Parlox.
On this page
AI agents now read, compare and buy on stores for the people they work for. Parlox shows a merchant which agents visit, what they read, where they get stuck and whether they buy. To do that it has to see two very different kinds of visitor.
Two kinds of agent
Fetchers. When someone asks ChatGPT or Claude about a product, the assistant fetches the page from its own servers. Search and training crawlers do the same. These requests name themselves in the User-Agent header (ChatGPT-User, Claude-User, PerplexityBot, GPTBot), but they never run JavaScript. A tag in the page cannot see them. Only the server that answered the request can.
Browser agents. Some agents drive a real browser: Gemini's agent, Claude's browser, Meta's Muse. They run JavaScript, and many send the same User-Agent as an ordinary copy of Chrome. Only signals inside the page tell them apart: how the browser is built (software graphics, a browser without a consumer brand) and how it behaves (clicks with no mouse movement before them).
| Visitor | Runs JavaScript | Who sees it |
|---|---|---|
| Fetcher (an assistant reading a page, a search or training crawler) | No | Your server only |
| Browser agent | Yes | The page, and your server |
| Person | Yes | The page, and your server |
Two parts, one gateway
Parlox therefore has two parts. Both send to the same gateway, https://gateway.parlox.io, and the dashboard at app.parlox.io shows them together.
- The browser part runs in your pages:
@parlox/browserfrom npm, or the pinned script tag where there is no build step. It counts page views, summarises how each page was used, records clicks and commerce events, and tells browser agents from people. Visits by automated browsers are also recorded as a masked replay. - The server part runs on your server:
@parlox/serverfrom npm, or one HTTP call from any other stack. For each page request that looks automated it sends one small report in the background, never delaying the response. Requests from people are filtered out on your server and never sent.
your pages @parlox/browser or the tag ──▶ POST /v1/b (public key, from your own domain)
your server @parlox/server or HTTP ──▶ POST /v1/s (secret key)
│
gateway.parlox.io ──▶ your dashboardThe browser part alone shows people and browser agents. The server part alone shows fetchers and crawlers. A store that installs both sees all three.
What a visit led to
The third piece is sales. Parlox never guesses what a click meant: a button that says "Buy" can add to the cart, open the checkout or place an order. Cart and checkout steps come from something that knows: your store's own cart API, your existing GA4 data layer, a call from your code, or Shopify's checkout events. The purchase of record comes from your server. See Cart and checkout events and Purchases from the server.
What you see
Each visit in the dashboard carries a label with its basis (verified, declared or inferred), the pages it read, the steps it reached, and one outcome derived from recorded facts, such as "Blocked by a CAPTCHA" or "Completed the purchase" (Visits and outcomes). Agents are compared with people on the same pages, and people who arrived from an AI assistant are counted apart (People sent by AI assistants).
Ready to install? Start with the Quick start, or see which install route fits your store.