How it works

We crawl your public pages once and keep the text as the agent's knowledge: product and service pages, policies, FAQ, shipping and returns, contact details. Nothing behind a login, nothing you have not already published.

There is a ceiling on how much we load per site. A very large catalogue may need a search-based setup rather than a flat text base. We check that before we quote you, not after.

Two small pieces go on your side: one <script> tag on the pages where you want the widget, and a short proxy route on your own server, about a page of code, which we write for you. The proxy holds the API key and tells the agent which of your customers is asking.

The key never reaches the browser. The widget carries no identity of its own and talks only to your server.

Someone opens the widget and asks a question. The agent answers from your content, writes the reply out as it goes, and keeps the thread for the rest of that browser session.

If it cannot help, or a customer asks for a person, it hands the conversation to your team with a short summary of what the customer wanted.

You can give the agent a few read-only lookups, for example order status or stock levels, so it can answer "where is my order" without a human. Each call goes through your server and is scoped to the customer asking, so no one can pull another customer's details.

We check your site for changes on a schedule you choose, as often as every 30 minutes. A full re-read only happens when a page you care about has actually changed, so the knowledge base moves when your site does.

Tone, the topics it covers, the widget's wording, a new lookup: you tell us and we make the change here. Your script tag stays exactly as it is.