Building an AI chatbot for your website: step by step

An AI chatbot on your website answers questions around the clock, takes load off your team and qualifies visitors — provided it answers from your real content, not from a model’s general training knowledge. This guide shows, in four steps, how to build and embed such an assistant with Kyros. You don’t need a developer team or a credit card.

Step 1: Create the assistant

You start with a single field: the assistant’s name. There is no template gallery and no type picker at creation — you enter a name, and Kyros automatically provisions a matching knowledge base and an embeddable chat widget in the background. That keeps the start lean: you decide later, in Studio, how the assistant sounds and answers.

Step 2: Connect your knowledge

An assistant is only as good as the knowledge it can draw on. In the knowledge base you store the content the chatbot should answer from. You have three ways to do this:

• Crawl URLs: point Kyros at your website, help pages or docs, and it reads the content in automatically.
• Upload documents: PDFs, manuals or exports from your CMS.
• Write golden answers: hand-written model answers for your most important questions — they take priority and make sure sensitive topics are answered exactly the way you want.

Grounded answers cite their sources: every claim gets a clickable footnote back to the original. If a piece of information is missing, the assistant doesn’t guess — it says it doesn’t know. A further benefit: a knowledge base is reusable and can be assigned to several assistants.

Step 3: Configure it in Studio

In Studio you give the assistant its character. Three areas matter most:

Model & personality

Through the model router you pick from several providers — with transparent per-message credit billing and automatic fallback if a provider goes down. In the personality you set tone, language and behaviour: matter-of-fact and brief, or friendly and detailed.

Moderation

Moderation screens incoming messages before they reach the model or get stored. Six categories are on by default — including the protection of personal data (PII). Further categories such as health, financial or legal advice, as well as jailbreak attempts, can be enabled optionally. Flagged messages are withheld beforehand.

Iterate safely: variations, playground, apply

Instead of experimenting live on your production assistant, you work with variations and the playground: you try settings, test them in a protected space, and only take a version live once it convinces you — with a single “Apply to assistant” action. That way a half-finished configuration never reaches your customers.

Step 4: Embed the widget

Finally, you put the assistant on your website. You embed the chat widget with a single line of code — a <script> snippet or an <iframe> for restrictive environments. It works with any CMS and any framework. Match colours, fonts, logo and launcher position to your brand, and use domain allow-listing to decide which sites may embed the widget. A GDPR-worded privacy panel already ships inside the widget.

What now?

That’s your chatbot live: grounded in your content, in your design and GDPR-compliantly hosted in the EU. Watch the chat logs over the first few days, add any missing golden answers, and refine the personality through new variations. You’ll find the details on plans and credits on the pricing page.