dAi Pro – all features at a glance

dAi Pro is the AI solution by Dreistein – chatbot, search, voice chat, autonomous agents and interfaces, all working exclusively with your own knowledge. dAi Pro comes in two editions sharing the same core – as a TYPO3 extension (dAi Pro-TYPO3) and as a standalone web application (dAi Pro-Web). This page lists every feature at a glance. For an introduction with use cases, see the dAi Pro product page.

dAi Pro brings content from many systems together in one knowledge base – without migration. The originals stay where they are. dAi Pro re-reads them regularly.

• Website crawler – ingests entire websites or sitemaps, respects robots.txt and converts pages into cleanly structured text, including headings, lists and tables.
• TYPO3 pages – pulls page content straight from the CMS, along with images and linked files.
• Files and folders – PDF, Word, Excel, CSV, text and images, uploaded individually or as a whole folder.
• Connectors to third-party systems – Confluence, Notion, SharePoint, Google Drive, Nextcloud, Zendesk, Freshdesk, email mailboxes (IMAP), SQL databases, CSV/Excel files and directory listings.
• Automatic sync – sources are synchronized regularly. New, changed and deleted content is detected and updated.
• Linked files – documents linked from web pages are discovered automatically, indexed and coupled to their originating page.
• On/off per document – any document can be deactivated without deleting it. It disappears from all answers immediately.
• Custom instructions per source – an indexing prompt controls how a source is prepared during ingestion.

Our standard – your editors should never have to fix anything by hand. dAi Pro reads documents so carefully that the answers are right.

• OCR for PDFs and scans – even scanned documents are read reliably.
• Figures from PDFs – charts and photos are extracted as separate image documents and described by AI, making diagrams searchable too.
• Table preparation – tables from PDFs and web pages are transformed into a shape the AI can quote correctly.
• Image understanding – images automatically receive an AI description. Matching images appear directly in chat answers.
• Multilingual PDFs – documents containing several language versions are detected and stored separately per language.
• Live progress – ingest runs show their progress in real time in the backend.

At the heart of dAi Pro is a chat that answers exclusively from your knowledge base – using two techniques that dAi Pro combines automatically.

• RAG (Retrieval-Augmented Generation) – a hybrid search combining vector similarity and full-text search finds the relevant passages and hands them to the language model.
• CAG (Cache-Augmented Generation) – thematic knowledge bundles sit pre-warmed in the language model’s cache, for particularly fast and inexpensive answers on frequent topics.
• Automatic routing – a classifier decides per question whether RAG or CAG answers.
• Strict grounding – what is not in the knowledge base does not get answered. There are no invented facts.
• Source citations that jump to the passage – every answer links the documents and pages it is based on, for connector content straight into the source system. Optionally the links jump right to the quoted passage, and for PDFs to the matching page.
• Answer cache – recurring questions are answered without another AI call.
• Assistant profiles – several assistants per collection, each with its own name, greeting, goal and slice of the knowledge base.
• Product finder – a guided dialogue mode leads visitors step by step to the right product.
• Tools (function calling) – the chat can call defined HTTP and SQL tools, for example to look up stock levels or prices live. Guard rules precisely limit what the AI may access. Identity-bound parameters – such as a signed-in visitor’s customer number – are filled in by the server; the AI never sees them.
• Human handover with context – at the visitor’s request, or when the chat cannot help, it offers to hand over to your team: as an e-mail message, as a callback with a preferred time, or as a ticket in your ticket system. The enquiry arrives with the full conversation and contact details, optionally with a copy to the visitor, and carries a reference number. Spam protection runs via Cloudflare Turnstile.
• File and photo upload – visitors can attach a photo or PDF, say a type plate or a manual, and ask questions about it. The file is only read for that conversation and never stored.
• Answer feedback – visitors rate answers thumbs up or down. The evaluation shows where the knowledge base needs sharpening.

The chat knows who is signed in — and answers personally. The identity never comes from the conversation itself, but always from a verified channel.

• Verified identity – the chat recognises TYPO3 frontend logins automatically. Any other website hands over a short-lived, signed token; invalid or expired tokens never break the chat, they merely make it anonymous.
• Personal greeting – the assistant addresses signed-in visitors by name.
• Personal data only through tools – order status, points balance or contract details come from your systems via function calling. Identity-bound parameters such as the customer number are invisible to the AI; the server fills them in from the login.
• Pass-through to your API – either with the visitor’s token or their user ID. Your system checks access itself; dAi Pro makes no decisions about other people’s data.
• Privacy built in – no customer accounts in dAi Pro, the chat log stores only a pseudonym, personal answers never enter the answer cache, and the conversation ends with the logout.
• Login at runtime – when someone signs in or out, the widget switches the conversation without reloading the page.

The chat does not just answer questions – it gets things done. Callback, ticket and appointment are confirmed actions: the assistant prepares a card with all the details, the visitor checks and confirms, and only then is the action carried out. Every action gets a reference number that visitor and team can both see.

• Callback – phone number and preferred time from your time windows. The request goes to your team by e-mail and, optionally, via API to your CRM or any other system – with field mapping, custom headers and the case number back into the chat.
• Ticket – created directly in your ticket system, with the ticket number shown in the chat right away: Zendesk, Freshdesk, Zammad, Jira Service Management and HubSpot are built in, any other system via HTTP interface. Without an integration, every ticket system accepts the request by e-mail.
• Appointment – either as a booking link (Calendly, HubSpot Meetings and any other service, with pre-filled contact details) or booked right in the chat: pick a free time, confirm, receive the invitation – with Cal.com, Microsoft Bookings, Microsoft 365 and Google Calendar.
• Recognised in conversation – “Please call me back”, “I want to open a ticket”, “I need an appointment”: the assistant recognises the wish in seven languages and opens the matching card.
• Signed-in visitors – name and e-mail address are already pre-filled.
• Safe and traceable – consent naming the recipient, only the confirmed details are transmitted, spam protection and rate limiting, no duplicate ticket on a double click, every action in the chat log with reference and case number.

• A real search, not a chat substitute – a dedicated search content element with hybrid technology (full-text + vector) also finds paraphrases, not just exact word matches.
• AI answer above the results – on request, the AI condenses the results into a direct answer.
• Suggestions and starter questions – guide visitors to their goal faster.

dAi Pro is multilingual from the ground up – in the frontend and in the backend.

• Knowledge base per language – documents carry their language. Visitors receive answers and sources in their language, with a clean fallback when a version is missing.
• Language detection – the language of every question is detected. For short messages, the conversation history helps.
• Automatic translation – missing language versions of documents can be generated automatically and kept up to date.
• Backend in seven languages – German, English, French, Italian, Dutch, Polish and Ukrainian.

• Speak instead of type – visitors ask questions via microphone and hear the answer, streamed fluidly while it is being generated.
• Selectable audio providers – Mistral (Voxtral), OpenAI or Gladia for speech recognition. Text-to-speech comes with selectable voices per language.
• Pronunciation care – technical terms and proper names can be corrected for speech output.

Agents monitor your knowledge base independently on a schedule and report by email – the decision always stays with you.

• Freshness guardian – finds outdated content, dead links and stale language versions. Unreachable material can be cleaned up automatically.
• Content quality – checks documents for contradictions, gaps and inconsistencies.
• FAQ generator – turns real user questions and your content into ready-made FAQ suggestions.
• Gap radar – analyzes chat logs and shows which questions your knowledge base could not answer.
• Ticket assistant – prepares support requests with draft answers from the knowledge base.
• Custom instruction – an agent with a brief you write yourself, for everything the other types do not cover.

The copilot supports your editors with five task types – strictly based on the knowledge base, with nothing made up.

• Drafting page copy from existing knowledge.
• Writing meta descriptions for search engines.
• Creating summaries of longer content.
• Preparing social media posts.
• Generating image alt texts for accessibility.

• Four providers – Mistral, OpenAI, Anthropic (Claude) and Google Gemini. They are freely combinable per collection, with chat and embedding models configured separately.
• The European option – with Mistral, processing stays entirely in Europe.
• Resilience – configurable fallback models step in when a provider is unavailable.
• Speed in view – our live RAG comparison shows how fast the models respond.

• TYPO3 content elements – chatbot and website search as ready-made content elements. Place them, pick a collection, done.
• Embed widget – one line of HTML embeds the chat into any website, including outside TYPO3.
• Design per collection – the chat widget’s colors and appearance can be customized per collection.
• REST API – a complete JSON API (/api/v1) with token authentication for projects, documents, chat, agents, users and more, including an OpenAPI description.
• Slack bot – dAi Pro-Web answers questions right in your Slack channels.
• Login integration – the embed widget accepts a signed identity token from your website (data-user-token), including login and logout at runtime.
• Two product lines, one core – dAi Pro-TYPO3 as an extension inside the CMS, dAi Pro-Web as a standalone application for every other environment.

AI assistants such as Claude or ChatGPT are increasingly visiting websites themselves. dAi Pro turns them into conversation partners. Instead of laboriously crawling pages, agents ask their question directly – and receive the grounded answer with sources.

• MCP server – every collection can act as a server speaking the Model Context Protocol. That is the standard AI agents use to connect tools. An “ask” tool answers questions from the knowledge base.
• Plain question endpoint – crawlers without MCP simply call a query URL. The answer comes back as machine-readable Schema.org JSON with a source list.
• Discoverable via llms.txt – both endpoints are announced in the site’s llms.txt. That is the place AI crawlers check first.
• With every safeguard – agent answers run through the same pipeline as the chat, with grounding, source citations and answer cache. Rate-limited per IP and switchable per collection.

• User management – users and groups with granular permissions per collection.
• Cost transparency – every AI call is logged with token usage and cost in euros, analyzable per collection and period, exportable.
• Chat log – every question and answer traceable, with language, origin and the technique used (RAG/CAG).
• Collection copies – documents can be copied in bulk to other collections, for example from staging to production.
• Setup assistant – a guided setup check verifies database, vector extension and API access step by step.
• Background operation – ingest queue, agents and CAG warm-up run automatically via scheduler or cron.

Since late June 2025 the German Accessibility Act (BFSG) obliges many websites to be accessible – and that includes their chatbot. The dAi Pro widget is built for it.

• Screen reader ready – the chat history is a live region. New answers are read out without stealing focus. While an answer is being written, a status line announces it.
• Clean streaming announcements – answers arriving word by word are announced only once complete. There is no stuttering in the screen reader.
• Correct pronunciation – every answer carries its language as an attribute. Screen readers switch pronunciation automatically, even in multilingual chats.
• Fully keyboard operable – visible focus indicators, focus management in dialogues, Escape to close, focus returned to the trigger.
• Reduced motion – animations respect the system setting for reduced motion.
• Labelled controls – every button and field carries labels in all seven chat languages.

• Your data stays with you – the knowledge base lives in your own PostgreSQL database, not in a third-party cloud index.
• GDPR-compliant – with European providers such as Mistral, processing stays entirely within the EU. Read more under GDPR-compliant AI.
• AI transparency – the chat identifies itself to visitors as an AI, as required by the EU AI Act.
• Signed-in users – identity is federated from your website; dAi Pro keeps no customer accounts and stores only a pseudonym in the log.

• TYPO3 12.4 to 14 – dAi Pro-TYPO3 runs on all current TYPO3 versions.
• PostgreSQL with pgvector – proven, open vector technology, cleanly separated from the TYPO3 database.
• Start for free – the free edition dAi offers the core (website crawler and RAG chat) as an open-source extension.

Want to see a feature in action? Get in touch – we will gladly show you dAi Pro working on your own content.