Scope: what your package includes

Three terms decide the package – here they are in plain words

The dAI Pro packages differ in three sizes: how many indexed records your knowledge base holds, how many data sources you connect, and how many agents work for you. This page explains all three, so you know where you will land before you start – not with the first invoice.

What counts as an indexed record?

Everything that appears as its own row in your document list – countable, at any time

A record is one entry in your collection's document list. You need not estimate anything and take nothing on trust: the figure is in the backend and you can read it yourself whenever you like.

What gets a row of its own:

  • Every web page – one URL, one record.
  • Every language version. The same page in German and English are two records, because each is prepared and searched separately.
  • Every file – a PDF, a Word document, a spreadsheet.
  • Every illustration we extract from a file. A data sheet with twelve drawings makes thirteen records: the file and its twelve images. Each image is captured by text recognition and can be found on its own – which is why it counts on its own.
  • Every row from a database query, every email, every Confluence article.

An example from our own operation: This website has around 75 pages and 2 PDF documents. Every page exists in two languages, the PDFs too, and the around 490 images on the pages count individually. So our own knowledge base counts around 650 records. That shows you: images and languages drive the figure – and in return the chat answers questions about illustrations and replies in both languages from the correct version.

In short: Do not ask how many pages you have, but how many rows end up in the list. We tell you the figure before you start – after a trial run across your sources.

What counts as a data source?

One entry in the source list – and how many you make of it is your call

A data source is one configured connection to a system. What matters is not how much sits behind it, but how many entries you create:

  • One website source can cover several domains. Put five domains into one list and that is one data source.
  • Prefer it tidier – say one source per domain, each with its own schedule and settings – then each one counts separately.
  • The same applies everywhere: a folder, a mailbox, a database query, a Confluence space – one entry each, one source each.

So you have the choice: bundle frugally or separate cleanly. Both are right; they simply cost differently.

Available among others: website indexing, TYPO3 pages, individual files and web links, folder scanning, CSV and Excel files, database queries, Confluence, Notion, SharePoint, Google Drive, Nextcloud, Zendesk, Freshdesk, email mailboxes and direct input.

Which four agents in the basic package?

You pick four – here are all of them, so you know what you are forgoing

Agents are the background workers of your knowledge base: they keep content current, check answers and report gaps. In the basic package you choose four; with the full scope they all run.

  • Freshness guardian. Keeps the knowledge base current – editorial changes immediately, the rest via sitemap and deep inspection. The one we recommend first: without it everything else goes stale.
  • Quality audit. Weekly report on real conversations: resolution rate, topics with a trend, an AI judge for answer quality. The second one we recommend.
  • Golden set. Check questions after every change to knowledge, prompt or model.
  • Content sanity. Finds expired deadlines, outdated prices and contradictions in your material.
  • Content gap radar. Evaluates real questions and delivers gap topics with sample questions.
  • FAQ generator. Turns unanswered visitor questions into publication-ready FAQ drafts.
  • Ticket assistant. Watches a support inbox and drafts answers from the collection's knowledge.
  • Free prompt. An instruction you write yourself, run on a schedule and filed as a report.

Our recommendation for the basic package: Almost everyone takes the freshness guardian and the quality audit – one keeps the knowledge current, the other shows whether the answers hold up. The two remaining slots depend on your plans: golden set if you change content often. Content gap radar and FAQ generator if you want to grow the site. Ticket assistant if support was the reason you came.

All agents in detail – with sample reports and screenshots.

How the chat gets onto your website

One line in the source code – whatever your website is built with

You need not rebuild anything and install nothing. The chat is added through an embed code: a single line you place once in your page template – the way you know it from analytics or map services.

This works in any system: WordPress, Shopware, Joomla, a hand-built portal or a static site. Your website needs to do nothing beyond serving HTML.

  • As a bubble in the bottom right – the default: a button that opens on click.
  • Or fixed inside the page. With data-mode="inline" and a target element the chat appears within your content instead of as a bubble – on a support page, for instance.
  • Several widgets side by side. Each embed can use its own assistant, its own language and even its own language model – via data-task, data-language and data-llm.
  • Adjustable appearance. Accent colour, title and greeting are set through further data- attributes, without touching any code.

Running TYPO3? Then you do not need the snippet at all: dAI Pro-TYPO3 comes with a content element you place like any other – plus a page chat you can switch on for the whole site with one click.

Einbettungs-Code html
<script src="https://ihre-instanz.de/embed.js"
        data-project="1"
        data-task="5"
        data-language="de"
        data-title="Ihr Assistent"></script>

When a package no longer fits

Limits are not traps

Knowledge bases grow. When your collection approaches the limit of your package, nothing surprising happens: you get a heads-up before it does, and we talk about moving up. Nothing is switched off automatically and nothing is quietly added to the invoice.

The same works the other way: if it turns out you need less, you move to the smaller tier.

And if you want to know in advance where you will land: we run a trial pass across your sources and give you the figure before you decide.