AI in TYPO3 14: writing, translation and custom tools.

TYPO3’s approach: interfaces without a fixed AI provider

TYPO3’s June 2025 concept article outlines an open architecture rather than a built-in AI provider. Workspaces, page trees, languages and user permissions should govern which content an AI tool can read or change.

The concept also explores MCP as a connection to AI services, rather than promising a core feature. A separate TYPO3 MCP Server extension is now available for AI agents to access the CMS. Our dedicated MCP article looks at how it can help prepare pages, content and media.

What TYPO3 14 provides for AI integrations

TYPO3 14 lets developers add translation methods to its translation workflow. An extension can send content to a translation service or AI model and return the result as a language version in TYPO3. Editors can prepare translations without moving between their CMS and a separate chat window.

The core supplies the workflow and the places where extensions can connect. A separate service produces the translation. Writing assistants, image analysis and other AI features also come through extensions or custom integrations: upgrading to TYPO3 14 does not install them automatically.

Five extensions for publishing, search and custom AI tools

Some tools work inside the text editor; others process entire records or provide a foundation for custom features. Before choosing an extension, we check the specific release against TYPO3 14 and the other extensions in the project.

AI Suite · AutoDudes

Draft, shorten and revise copy, translate content and complete page and image metadata. Reusable instructions can define a house style, while batch processing helps with larger content collections. Image generation and suggested page structures are also part of its feature set.

AI Writer · ok_ai_writer

A writing assistant inside CKEditor: editors describe a task, review the proposed text and refine it through a conversation. Translations preserve the HTML structure. It connects to OpenAI or Azure OpenAI; the documentation also lists support for TYPO3 14.

DeepL Translate · web-vision

Translates pages, content elements and other records through DeepL, extending support beyond an individual text field. Its package name is deepltranslate_core, but it remains a separately installed extension rather than part of the TYPO3 core.

T3AS · T3Planet

Adds AI-generated answers to website search, supporting Solr, ke_search and indexed_search. Knowledge sources can include pages, news and PDFs. It is a candidate for answering questions about existing content; the project still needs to check source selection, access rules and answer quality.

TYPO3 LLM · Netresearch

A foundation for custom AI extensions with centrally configured providers and models, including OpenAI, Claude, Gemini and Ollama. Custom text-processing, image-analysis or semantic-search features can build on it without developing every connection separately.

What this could look like in everyday editorial work

Three examples of workflows we would assemble and configure using these capabilities:

Turn a specialist article into useful shorter versions

An approved article becomes the source for a teaser, a newsletter introduction and a short page description. Length, tone and terminology are specified. Editors compare the suggestions and select the appropriate version.

Introduce a product to additional markets

The description, technical details and image copy are prepared for the required languages. Product names and specialist terms follow defined conventions. Each country’s editors check the wording and local requirements before publication.

Complete the metadata of an existing image library

Images without useful titles or descriptions receive suggestions for review. For alternative text, we also consider how the image is used on its particular page. Image analysis alone does not have that context.

Build AI tools around your TYPO3 content

A knowledge assistant could search selected TYPO3 content and answer questions with links to its sources. Another tool could read incoming product information and suggest values for the relevant fields. We develop the connection between the content model, the AI service and the interface people use.

nr-llm handles communication with the models. The application enforces access to content and search results and provides a way to transfer reviewed suggestions into existing records.

Agree on data use, approvals and costs

We start with one task and real examples from the editorial team. How much effort does it take to review the output? Does it preserve specialist terminology? What happens when information is missing? Those answers matter more than a convincing demo.

Before connecting a service, we agree which content it may receive, who can use the tools and who approves the results. We also set cost limits, handle service failures and check the integration with existing content types and workspaces. Generated content needs a reviewable draft stage before publication.