Your website is already the knowledge base

Open the setup screen of almost any small-business chat tool and you'll find the same thing: a long form asking what you do, where you are, when you're open, what you charge, and how someone books you. It's a reasonable-looking screen. It is also the point at which most people close the tab.

Not because the questions are hard. Because they've all been answered already, in public, on the website the visitor is currently standing on. Being asked to type your opening hours into a second system — knowing you'll have to remember to update both when the Saturday roster changes — is a fair preview of how much work the thing is going to be forever.

Read the site instead

So the setup is a single field: your web address. We fetch your pages the way a visitor would, pull the text out, and use it to answer questions. A typical small-business site — a dozen pages, a services list, a contact page — takes a few seconds.

Two things about doing it this way turn out to matter more than the speed. The first is that you have to be shown what was found. Chunks of text and vector embeddings are the right machinery, but they are not something you can check, and “are my hours right?” is the question that decides whether the widget stays switched on. So the crawl is also summarised into a handful of topics in plain English — hours, contact, services, pricing, booking — that you can read in under a minute and correct in place.

The second is that the footer is the most valuable part of the page. Stripping boilerplate is the obvious first move when you're extracting text, and on small-business sites it's exactly wrong: the opening hours, the phone number, the address and the email almost always live down there. Take the footer out and you have built a chat agent that cannot answer the most common question it will ever be asked.

What it can't read

It reads text. A site that renders its content as JavaScript, or as images of text, will come back with very little — and you'll see that plainly rather than finding out from a customer. The fix is the same either way: type the important things into the topics yourself. The agent treats what you wrote and what it read exactly alike.

And when the site changes

Re-reading your website is a button, not a schedule, and anything you've edited by hand survives it — see why your edits outlive the crawler. If your site now says something different to what you wrote, you're shown both and you pick.