Website chatbot

What it does

• Scripted flows built in Studio: greetings, buttons, questions and lead capture.
• AI answers from your site’s own published content, paid for with your plan’s AI credits. When no credits are left, the bot gives its fallback message.
• Appearance and triggers: colours, launcher image and when it opens, set per site in Studio.
• Turnstile protection when the conversation looks automated.
• Leads: details a visitor shares in a lead step arrive in Forms & leads → Leads, like form enquiries.

There is no live hand-over to a human agent yet.

On the live website

Chatbots travel to Production inside releases. When Production is on and a signed release is live, the released chatbot answers real visitors, using the knowledge from the pages in that release (up to 5,000 passages). Development changes reach visitors only with the next approved release.

While Production is off, the chatbot addresses answer 404 CP_NOT_FOUND; with Production on but no live release, 403 CP_PRODUCTION_LOCKED.

Add it to a Next.js site

1. Render the widget once in your root layout. It needs no token: the browser talks to Atlas directly, and Atlas only accepts your registered website origin.

   app/atlas-chat.tsx

   'use client';
   import { AtlasChat } from '@atlas/chat-widget';
   import '@atlas/chat-widget/styles.css';
   import { usePathname } from 'next/navigation';
   
   export function SiteChat() {
     const path = usePathname();
     return (
       <AtlasChat
         apiOrigin="https://delivery.atlas.example"
         environmentId="env_XXXXXXXX"
         botKey="support"
         path={path}
       />
     );
   }

2. Allow Atlas in your Content-Security-Policy: add the Atlas API origin to connect-src and img-src.

3. Use the Production environment id in your live deployment, and the Development one everywhere else.

The widget package is distributed to connected customers directly; it is not on the public npm registry yet. Without React, use the plain client createChatClient({ apiOrigin, environmentId, botKey }), which offers bot(), start(path) and send(…).

Privacy

• Conversations are tied to a random secret kept in the visitor’s session storage. There are no cookies.
• Lead details are stored only when the visitor submits them in a lead step.
• Rate limits apply per IP address and per conversation; see Limits.

What to read next

Work with leadsBrowser endpoints