Agent visits: the request log
Every recorded request, newest first — what each column means, how the filters work, and why Googlebot shows up as a crawler.
Agent visits is the raw log: one row per recorded request, newest first, 50 rows per page (use “Older →” to page back, “← Back to latest” to return). With more than one store, pick the store with the switcher pills at the top.
The columns
| Column | What it means |
|---|---|
| Time (UTC) | When the request happened, in UTC (not your local time zone). |
| Kind | Who asked: agent (an AI assistant or its fetcher), human (a regular browser), crawler (a search engine like Googlebot), or protocol (a commerce-protocol client probing endpoints like /mcp). |
| Platform | Which agent platform, when identifiable: chatgpt, claude, claude-code, opencode, perplexity, gemini, googlebot, bingbot — or unknown. |
| Verified | Whether the agent cryptographically proved its identity (Web Bot Auth signatures). Verification is live in the gateway, but it can only say “yes” once trusted platform keys are on file — the shipped key registry is deliberately empty until real published keys are pinned — and the platform actually signs its requests. So expect “no” on almost every row today; that is normal, not an error. |
| Served | What answered: markdown (the gateway served the agent text), html-variant (pass-through — your normal HTML page; the “variant” wording anticipates a future feature, today it simply means your standard page), or protocol. |
| URL | The page on your store that was requested (hover for the full URL). |
| Render | How long the gateway spent producing the response, in milliseconds. Markdown rendering is pure computation designed to a sub-50ms budget. |
Filters
Two chip rows above the table filter the log: Kind (all / agent / human / protocol / crawler) and Platform (all, plus each platform). Filters combine, and “No visits match these filters” offers a one-click “Clear filters”.
Why Googlebot is a crawler that gets HTML
Googlebot (and bingbot) appear with kind crawler and Served html-variant — never markdown. This is deliberate: serving search engines different content than humans see is called cloaking, and it can get a store penalized. The gateway therefore always gives known search crawlers your canonical HTML, even if they ask for markdown. What agents get is the same substance in a different format — so there is nothing to hide from Google in the first place.