How we Built a Fully Open-Source AI Chatbot for TYPO3

Most chatbot solutions are based on external platforms. In practice, this often means:

• Data is processed outside your own infrastructure
• Vendor lock-in and dependency on a single provider
• Ongoing usage and subscription fees
• Limited control over language models and data storage

At undkonsorten, we needed a solution that can be deployed in environments with high data protection standards. Running an AI chatbot within TYPO3 has many advantages:

• Open source instead of SaaS
• Data storage directly in TYPO3
• Free choice of language models (LLM)
• No additional platform and payments required
• Full control of knowledge through internal retrieval-augmented generation (RAG)

Since chat histories are stored directly in TYPO3, the data remains entirely under the control of the website operator. Furthermore, chats can be deleted automatically with a scheduler task and existing TYPO3 content can be integrated as a knowledge base.

EasyChat is built entirely on open-source components. DeepChat serves as the chat frontend and provides the user interface. TYPO3 and Symfony AI process user queries and manage key tasks, including:

• Communication with the LLM
• Storage of chat data
• Consent management
• Request processing via TYPO3 Reactions
• RAG/Indexing with the new TYPO3 Indexer Extension
• Qdrant support as vector database

EasyChat is not locked to a specific provider. It is designed to be model-agnostic. Currently, we use the open-source model gpt-oss-120b (via AI hosting in Germany by TYPO3 partner Mittwald). However, you can also connect models like Mistral or ChatGPT directly if desired.

Luckily TYPO3 is already equipped out-of-the-box with many features we needed for our chatbot application:

• Admin Interface / Backend Modules
• Reactions to receive and handle incoming chat requests
• Scheduler Tasks for automated data cleanup and deletion
• Fluid Templating Engine for customized chatbot design and theming overrides
• Content (aka. knowledge) management: You might already have all the knowledge the chatbot would need as TYPO3 content
• RAG-Indexing solutions

Instead of introducing additional third-party systems, we leverage the existing TYPO3 infrastructure.

The extension of course is open source and we warmly invite you to contribute – so it can grow and become a standard TYPO3 chatbot extension.

• RAG Improvement — At the moment we are planning on improving the RAG feature so that we can index different kinds of content (tt_content, FAQs) and add it to the knowledge database.
• Vector Databases — At the moment we support Qdrant, but we will soon be supporting more and more databases like PostgreSQL or Redis.
• Answer Rating — Users can rate the answers so that the knowledge base can be improved.
• Conversation History — Improved insights of what the clients are actually asking or want to know from the bot in the TYPO3 Backend.

And many more.

• Demo page
• EasyChat in the TYPO3 Extension Repository
• EasyChat on GitHub
• Documentation