Answers about how this works

A web address. You add your site, HootyBot reads the public pages it is allowed to read, and the assistant can answer from them. There is nothing to install on your server and nothing to upload.

It depends on how many pages you have. HootyBot fetches one page at a time per site with a pause between them, so it is deliberately unhurried — it is reading your site, not stress-testing it. Pages appear in the console as they are read, so you can watch it fill up rather than wait for a finish.

For the chat widget, no: it is one script tag pasted into your page, and the console writes the line for you. The search module needs someone to decide where on the page the field goes, which is either a CSS selector or simply where you paste the snippet.

Public HTML pages that robots.txt allows. It never submits a form, never logs in, and never touches a cart or a checkout. Anything it is not allowed to fetch never reaches the index.

HootyBot, the crawler

Two lines in robots.txt. It fetches /robots.txt before anything else and follows it, including * and $ patterns, and it honours Crawl-delay. Every request carries its name and an address you can write to, so it is never an unexplained line in your logs.

The robots.txt rules

No. The pages HootyBot reads are used to answer questions about your site, and for nothing else.

If you ask it to. Each source can re-read itself daily or weekly, or stay off and only run when you start it. A page that has changed replaces its old text in the index.

The chat widget floats in a corner and holds a conversation. The search module sits where you put it in the page, takes a query, and returns ranked pages of your own site — clicking one goes there. It has no conversation at all. They read the same index, and you can run both.

Both modules in detail

The key in the script tag is public — the browser has to have it. What grants access is the list of domains you allow: the server checks the browser-set Origin against that list and refuses anything else. If a key does turn up somewhere it should not, rotating it kills the old snippet immediately.

Both modules render inside a closed shadow root, which isolates styles in both directions: your rules cannot restyle them, and their rules cannot leak onto your checkout. They carry no framework and add nothing to the page but one namespace.

Yes. Both are built for a narrow screen as much as a wide one — the search results reflow to a single column and the chat takes the full width rather than staying a small floating panel.

It says it could not find that on this site. The gap is not filled with something plausible. Those moments are worth reading: the conversations view shows you what your visitors asked that your pages did not cover.

The one the question was asked in, even when your pages are written in another. Someone can ask in German about an English page and get a German answer.

Yes, on Pro and Max. You take over a conversation from the console and type to the visitor directly; they stay in the same window and the assistant stops answering until you hand it back. The Free plan does not include taking over.

What each plan includes

The questions your assistant answers, the pages held in your index, and the tokens spent reading and answering. All of it is visible in the analytics view, filtered by date, agent and API key.

Plans and limits

On a paid plan, nothing breaks: answers keep coming and the extra is charged against your prepaid balance. On Free, answering stops until the month resets. Nothing is ever charged that the balance has not already covered, so there is no invoice arriving later with a number you did not expect.

Yes. A downgrade or cancellation takes effect at the end of the period you have already paid for, not the moment you click — and you can undo it at any point before then, from the same screen.