Your first chat in 30 minutes

Purpose

This chapter walks through the whole path once: create a collection, set up the website as a data source, index, process, ask in the test chat, embed. Afterwards you know every page you will need later. The following chapters go deeper into the individual steps.

You need a dAI Pro instance with at least one configured language model configuration (Chapter 24) and a website that is reachable from the instance.

The start page

After signing in you see the list of all collections you have access to. It is the starting point for everything else.

Column	Meaning
Title	The name of the collection. A click opens its collection view.
Schema	The technical name of the collection in the database. It is derived from the title on creation and stays fixed afterwards.
Languages	The languages the collection holds content for and answers in.
Updated	When the collection was last worked on.
Action	Opens the collection or deletes it: after confirmation, for good and including the index.

New collection sits at the top. Operators of the instance additionally see the list of demo collections below, with the result of the last self-test (Chapter 26).

Step 1 · Create a collection

On the start page, click New collection.

The “New collection” form with title, languages, type and language models

Field	What you enter
Title	The name of the collection. The technical schema name is derived from it and cannot be changed later.
Languages	The languages of your content, separated by commas, for example `de,en`. Determines the language selection in the chat and the answer language.
Type	“Chatbot” for the chat, “Website search” if the collection is only to serve the search. Empty = chatbot.
Chat / answer LLM	The language model that answers.
Embedding LLM	The model that creates the index. Cannot be changed after the first indexing.
Topic filter threshold	How strictly off-topic questions are rejected. The default is almost always right.

After “Create collection” you land in the still empty collection view.

Step 2 · Set up a data source

Open Data sources in the collection rail and, at the bottom under “Add data source”, choose the type Website indexing. After “Add”, the data source's form opens.

The form of the website data source with sources, crawl depth and content types

Enter a Label; it appears as a tab in the collection view.
Under “Sources & order”, click Crawl domain and enter the start address of the website. From there, dAI Pro follows all links within the domain.
Set Max. pages to a number that suits the website. It limits the run.
Leave the remaining fields at their defaults and click Save.

Step 3 · Start indexing

Back in the collection view (Collection in the rail), expand the Indexing box and click Start indexing.

The “Indexing” box with the “Start indexing” button

The run lands as a job in the list below. A background worker picks it up within a minute; the page shows the progress without reloading. With 16 pages, the run is finished within a few minutes. The pages are then listed under Crawl results with the status “crawled”: fetched, but not yet in the index.

Step 4 · Process

In the list, choose Select all and click Process. dAI Pro reads every page, splits it into sections and computes the index. The status changes to “indexed”, and the number Chunks (collection) at the top rises.

The document list with selection and the “Process” button — Without this step the chat knows nothing: crawling fetches the pages, processing makes them searchable.

Step 5 · Ask in the test chat

Open Chat in the rail. The test chat at the top answers from the index. Ask three or four questions whose answers you know and check the source references.

The Musterfirma test chat with greeting and starter questions

When you open this page for the first time, dAI Pro has already pulled suggestions for the header text, greeting, help text and the links to the privacy policy, legal notice and contact page from the website. All of it can be changed (Chapter 8).

Step 6 · Embed

At the end of the chat page, copy the embed code, insert it into your website and enter the website under Edit → Access and visitor features → Allowed websites. Chapter 8 describes all settings of the chat, Chapter 9 the search.

What comes next

Set the main prompt: role, tone, formal or informal address (Chapter 7).
Switch on human handover so that unanswered questions reach you (Chapter 12).
Create a freshness guard that checks the website for changes at regular intervals (Chapter 15).