Build your knowledge base from your website

What this does

Instead of writing your knowledge base by hand, you can point Chatonio at your own website. It reads your pages, splits them into short, focused articles, picks up any question-and-answer pairs you already publish, and turns them into real knowledge-base entries the AI can use straight away.

It is the same reader behind the instant demo, without the time limit — so it reads more pages, follows the links you give it, and works on the project you are already signed in to.

Where to find it

Go to Admin → Projects → [Your Project] → Knowledge Base → Articles and click Build from your website. You need to be an admin or manager on the project.

What you give it

• Your website — we start from this address and follow the pages most likely to answer support questions: FAQ, pricing, delivery and returns, instructions, terms, contact.
• Extra pages (optional) — paste specific links, one per line. These are always read; the automatic page-picking only fills whatever room is left.
• Anything the assistant should know (optional) — a few sentences in your own words, for example “We are a small hand-made shop. Never promise a delivery date.” This shapes the assistant’s instructions. It does not change which pages we read.
• Pages to read — how many pages to fetch. More pages means a fuller knowledge base and a slightly longer run.

Add, or replace

There are two modes, and the difference matters.

• Add to the knowledge base — nothing existing is removed. Always available.
• Replace what was generated before — clears the entries a previous run created, then writes fresh ones. This is offered only when nothing in your knowledge base was written by a person.

That rule is deliberate and it is enforced, not just advised: anything you or your team wrote is never deleted by this feature, in any mode. As soon as your project contains a single hand-written article or FAQ entry, replacing is switched off and you are shown how many of your own entries are protecting it.

The most common case for replacing is a project that started as an instant demo and was carried over at sign-up — everything in it was generated, so you can re-read your site as often as you like and get a clean result each time.

Visibility, and why entries are published straight away

New entries are Internal by default: the AI uses them immediately, and they do not appear on your public helpdesk. You can choose Public instead if you want them on the helpdesk too, and you can change any individual article's visibility afterwards.

Entries are always created as published rather than as drafts. That is not a shortcut — an unpublished article is not indexed for the AI, so a “review it first” draft pile would leave you with a knowledge base your assistant cannot actually use. Publishing them Internal gives you the benefit right away with nothing exposed publicly.

Everything a run writes is filed under a single From your website category, so reviewing, editing or clearing it out later is one filter away.

The assistant's instructions

Alongside the knowledge base, the run drafts a set of instructions for your assistant: who it is, what your business does, the facts it should know, the words your site uses, and what it must never claim. Every factual line is checked against the text actually found on your pages, and anything that cannot be traced back to your site is dropped rather than guessed at.

Those instructions are only applied if you tick the box. If your project already has instructions, the box starts unticked — we will not overwrite something you wrote. Either way the draft is saved with the run, so you can read it and copy the parts you like without paying for a second run.

What to do afterwards

1. Skim the new articles in the From your website category. They are short on purpose — one topic each — which is what makes the AI's answers accurate.
2. Delete anything that is not useful (menus, legal boilerplate you would rather it did not quote).
3. Switch the ones your customers should be able to read to Public.
4. Try the assistant. If an answer is thin, the fix is usually one short article about that topic.

When a site cannot be read

The run tells you exactly which pages it read and which it skipped, and why. The usual reasons:

• Blocked by robots.txt — we respect it. Allow ChatonioDemoBot in your robots.txt and run it again.
• Not enough readable text — pages that are almost entirely images, forms or interface. We try once to render a JavaScript-heavy homepage, but a site that renders everything client-side may still come back thin.
• Could not be reached — a login wall, a bot challenge, or a page that is simply down.

Pages behind a login are never readable, since we visit your site the same way an anonymous visitor does.

Cost and limits

Reading your site costs nothing. Writing the assistant's instructions is one AI call, billed as normal AI usage against the project's balance — a fraction of a cent on the default model. There is no plan requirement; the feature is available on every plan.

One run at a time per project, and a handful of runs per day. If a run is going you can close the window — it keeps going, and reopening shows you where it got to.