AIvataryBot
If you see AIvataryBot in your logs, someone using AIvatary pointed it at your site as a knowledge source for their agent. This page says what the bot does, how fast it goes and how to stop it.
How it identifies itself
Every request carries this header, with no exceptions and no variants:
AIvataryBot/1.0 (+https://aivatary.com/bot)We do not pose as a browser and we do not spread across several names. A rule you write against this name genuinely applies to us.
How to block it
With an ordinary robots.txt entry. We read that file before the first request to a site and honour every Disallow directive:
User-agent: AIvataryBot
Disallow: /You can also close off a single part of the site, for example Disallow: /admin/. The block takes effect from the next scan.
What load it brings
A scan is not a race. Pages go one after another, never in parallel.
| Interval between requests | At least one second. If your robots.txt declares a Crawl-delay, we take your value where it is longer, up to ten seconds. |
| Concurrency | None. One request at a time against one site. |
| Pages per scan | Fifty at most, within three clicks of the starting address. |
| Scope | The same domain and the same protocol as the starting address. Subdomains are out of scope. |
| Download size | Up to two megabytes per page, with a ten second timeout. Images, video and archives are skipped entirely. |
What the bot does not do
It does not log in or submit forms
It sends GET requests only and attaches no authentication headers to them. It will not click send, will not add to a cart and will not create an account.
It does not execute JavaScript
It reads the HTML your server returned. If content arrives in the browser instead, the bot will not see it and will report exactly that to the person who ordered the scan.
It does not work around protections
It does not solve CAPTCHAs, does not change its name to slip past a rule and does not retry from another address. A protection wall ends the matter.
It does not train models on your content
The text serves only to answer questions inside the instance of the customer who ordered the scan. Language models come from commercial interfaces in no training mode.
It does not roam
We run no index of the web. The bot starts when a person in the AIvatary panel enters one specific address, and finishes once it has walked that one site.
It publishes nothing on its own
Collected content waits as draft material until a person approves it. Without that decision the agent will not quote a single sentence.
Questions or complaints
Write to contact@aivatary.com. Give us the domain and the rough time of the log entries, and we will find which scan it was and stop it if you ask us to.
We keep no fixed list of IP addresses the bot goes out from, so we do not ask to be allowlisted by address. Recognise us by the name in the User-Agent header.