Chatbot for TYPO3: options, extensions and integration

TYPO3 websites are ideal candidates for an AI chatbot: the content is structured, maintained in several languages and protected by permissions – exactly what a knowledge-based assistant needs. This page shows the three routes to adding a chatbot to TYPO3, explains what the free extension dAi from the TYPO3 Extension Repository can do and where dAi Pro goes further, and names the pitfalls we have met in projects. If you first want to place TYPO3 itself, the basics are on What is TYPO3?

Why TYPO3 websites are good chatbot candidates

  • Structured content – page tree, content elements, heading hierarchy: a chatbot can form clean knowledge sections from these instead of guessing through HTML soup.
  • Multilingualism built in – language versions are cleanly separated; the bot can keep a separate knowledge base per language and answer in the visitor's language.
  • Permissions and visibility – hidden, time-controlled or group-restricted pages are marked in the CMS. A real integration respects that; an external crawler only sees what is public.
  • Changes are known – TYPO3 knows when a page was saved. An integrated bot learns immediately instead of waiting for the next crawl.
  • Files with metadata – PDFs and images live in the file management with title, description and language.

Route 1: embedding an external widget

The fastest route: a chatbot vendor crawls your website, you add one line of JavaScript to the page template. Advantage: no installation in the CMS, live in minutes. Disadvantages: the bot only sees the public HTML view, learns about changes only at the next crawl, knows neither permissions nor editors, and the vendor dictates data protection and model choice. Often sufficient for a small website with few pages – if the vendor offers grounding and European processing. What to look for when choosing is on Comparing AI chatbot vendors.

Route 2: the free extension dAi from the TER

dAi is our free TYPO3 extension, installable via Composer or from the Extension Repository. It reads your own TYPO3 pages, stores embeddings in a PostgreSQL database with pgvector and answers visitor questions via RAG or CAG with the language model of your choice – Mistral, OpenAI, Anthropic or Gemini. The chatbot goes onto any page as a content element; a backend module shows collections, documents and usage. Requirements: TYPO3 12.4 to 14, PHP 8.2, PostgreSQL with pgvector and an API key with the model provider. The manual leads in eight steps from installation to a running chatbot – allow an hour, most of it for the database.

Route 3: deep integration with dAi Pro

Where dAi stops, dAi Pro begins – as a TYPO3 extension with the same operating logic but considerably more reach:

  • More sources – besides TYPO3 pages also PDFs and Office files from the file management, server folders, database queries, Confluence, SharePoint, Nextcloud, ticket systems, mailboxes – 16 data source types, combinable in one knowledge base.
  • Changes immediately – when an editor saves a page, TYPO3 reports it to the knowledge base; a freshness monitor additionally checks all sources and clears deleted content.
  • Website search – the same knowledge base as hybrid search with result lists, also as a content element.
  • Personal answers – logged-in visitors get answers from content only their group may see, with a server-side signed identity.
  • Voice chat, handover, actions – voice input and read-aloud, handover to humans including history, callback, appointment and ticket straight from the conversation.
  • Agents and editorial copilot – FAQ suggestions from real visitor questions, gap radar, quality checks, text suggestions in the backend.

The complete list is in the feature overview. dAi Pro is also available as a stand-alone web application for websites without TYPO3.

Embedding on the website: content element, hero, modal, embed

In TYPO3 the chatbot is placed like any other content element – on a contact page, a product page, as a full-screen assistant or as a solution finder right in the hero of the start page. A click trigger in the page header opens it as a modal on every page. For sites outside TYPO3 or for partner websites, one embed line is enough. Design and texts can be adapted per collection; the chat window is developed to WCAG 2.2 and runs without cookies. You can see an example on this website: the assistant at the top right knows exactly these pages – and names the source for every answer.

Content: what the bot should read – and what not

Not every TYPO3 page belongs in the knowledge base. A selection has proven itself: service and product pages, FAQ, instructions, data sheets, knowledge pages – yes. Legal notice, login pages, form pages without content, archived news from 2014 – no. In dAi and dAi Pro you decide document by document which pages and files the bot knows; system folders, hidden pages and pages with noindex stay out. For editors a look at the extraction pays off: headings are carried into every knowledge section, tables are read out structured, images described – the cleaner the page is maintained, the better the answers.

Multilingual TYPO3 websites

TYPO3 separates language versions cleanly – the chatbot should too. dAi and dAi Pro keep a separate knowledge base per language, detect the visitor's language and answer in it; welcome message, starter questions and help texts are maintained per language. Important with files: a multilingual PDF is split into language sections so that a French visitor is not quoted the German page. Anyone who has translated a language only partially should choose the fallback deliberately – better an honest answer in the source language than an invented one in the target language.

Data protection in the TYPO3 environment

An integrated chatbot runs on your TYPO3 server: knowledge base and logs stay there, only the passages needed for the answer go to the language model – by default Mistral in France. The chat window sets no cookies, a privacy notice is one click away, conversation logs are deleted automatically after the period. What you as operator have to regulate beyond that – privacy policy, data processing agreement, labelling under the AI Act – is summarised in the guide Chatbot and GDPR.

Operation: scheduler, worker and freshness

Reading, embedding and updating run in the background – in TYPO3 via the scheduler or a cron job that works through the queue every minute. After setup the bot needs no manual work: TYPO3 reports page changes directly, the freshness monitor checks external sources at intervals, dead content is cleared. In the backend you see progress, errors and usage per collection; the usage log shows every AI call with its cost in euros.

Typical pitfalls

  • pgvector missing – the vector database needs PostgreSQL with the pgvector extension; many hosting packages only offer MySQL. A small PostgreSQL instance alongside is enough.
  • Scheduler not running – without a cron job the queue stalls; the backend shows it, but nobody looks.
  • Too much in the knowledge base – whoever reads in everything gets fuzzy answers. Better to curate.
  • Cache after changes – page cache and answer cache are two different things; after content changes the knowledge base updates, old cached answers expire after the period.
  • Permissions overlooked – a page visible only to one frontend group must not end up in the public collection.
  • No grounding test – ask trick questions before going live; the bot must be able to say "I do not know".

Chatbot or website search – or both?

A chatbot answers concrete questions; a search helps with browsing. Both can run on the same knowledge base: dAi Pro brings a hybrid website search as a content element that combines vector and full-text search and thus also finds article numbers and technical terms. Many operators start with the chat on service pages and add the search later – or the other way round. If you want to extend TYPO3 with AI as a whole, further building blocks are on AI for TYPO3.

Frequently asked questions about chatbots in TYPO3

Yes: dAi is a free TYPO3 extension from the Extension Repository that answers questions about your own TYPO3 pages via RAG or CAG. It needs PostgreSQL with pgvector and an API key with the language model provider.

dAi and dAi Pro run with TYPO3 12.4, 13.4 and 14 on PHP 8.2 or newer. Integration is via Composer; a TER package is available for non-Composer installations.

With dAi Pro, yes – including tables, scanned pages via OCR and figures with descriptions. Files linked on pages are read automatically and coupled to the page.

With the integrated variant, immediately: when a page is saved, TYPO3 reports the change to the knowledge base and the update runs in the background. External widgets see changes only at the next crawl.

No. dAi Pro runs in a normal web hosting environment with its own database; on request we operate it on our servers in Germany or entirely on your infrastructure.

Yes. Each language has its own knowledge base with its own prompts and welcome messages; the bot detects the visitor's language and answers in it. Multilingual PDFs are split into language sections.

Yes: a free demo with up to 50 pages of your website is ready in about 15 minutes – without installation, with a data processing agreement, deleted automatically after seven days.