Skip to content
Language
DE EN FR
Start a project
Start a project Service desk

AI as part of the platform architecture

Structured data, interfaces and fixed connection points. Your AI tools access them, and we use AI within the stack, from text suggestions to data normalisation.

Book a call What the AI takes on

The illustration shows the stack as three layers. The AI sits in the middle layer as a highlighted point; its paths reach into the content above and the data below, and connections lead outwards to three AI tools.

Switzerlandas the location, operated to ISO/IEC 27001
MCP/A2Aopen standard for your AI tool
ModelsWe use open-source models
SchemasDomain logic lives in the schema, not in the model
Stance

Where we use AI and where we don’t

By now, every platform claims ‘AI integrated’. The difference lies in what it is used for, and what it is deliberately not used for.

What we use AI for

  • Suggesting translations and flagging when the source has changed
  • Generating alt text, keywords and image variants
  • Revising texts in each client’s tone of voice
  • Answering questions on how to use the system and researching the platform’s own content
  • Answering visitors’ questions in the chat widget from the published content
  • Asking follow-up questions and sharpening a request in dialogue until the goal is clear
  • Suggesting topic-relevant internal links
  • Reading documents and classifying their content by subject

What we don’t use it for

  • Publishing unchecked: every AI result remains a suggestion
  • Writing legal texts, prices or binding commitments
  • Adding facts that appear nowhere on the website
  • Answers from the open internet: the assistant and chat widget stay within the platform’s own content

Your AI tool works with your platform

Your own AI tool, for example Claude, ChatGPT or Copilot, runs outside the platform, whether a provider supplies it or you develop it yourself. It connects to the platform via open standards such as MCP and A2A.

Content and data, your AI tool

Reading pages and creating drafts, but also querying and analysing data: your AI tool works not only with the website’s texts but with everything you release to it in your platform.

  • Separate permissions: in a regulated environment, an AI tool needs clear boundaries. You authenticate it and decide what it may do and which content and data it works with. Its permissions are separate from the user accounts and can be configured individually; it inherits nothing from your account.
  • A suggestion before every change: the AI tool suggests, you accept or discard, and every change can be undone. Publishing stays with the editorial team.
  • Your environment: your AI tool runs on your side, not in the platform. What it does with the content it reads is up to you. What it may see and change is decided by the platform.
An AI tool is asked to prepare the autumn promotion with the best-selling milling cutters of the third quarter. It queries the sales data and creates a draft in the cloudtec Stack. The editor shows the changes as a preview with the buttons ‘Discard’ and ‘Accept’. Bottom left, a note says that Anna Meier authenticated the AI tool and that its permission covers drafts.
In the platform

An agent with a clear brief

We also develop agents that work directly in the platform. Each one is a separate instance with a clearly defined task. We decide which functions it uses in which processes.

An agent with a fixed process

An agent regularly carries out a clearly defined task, for example a report based on the platform’s data. Structure and content are predefined. The platform supplies the figures and facts; the agent adds only what needs interpretation: a summary or an assessment.

It uses external sources only once we have checked and approved them. A guard monitors the network traffic between source and agent. An independent logging instance records every reasoning step and every action. This means you can trace at any time where a statement comes from.

We choose the language model behind it with you based on the task, your business requirements and data protection, because the largest is not always the most economical. Which functions the agent uses does not depend on it.

What AI takes over in everyday editorial work

The features start where editorial teams lose time: shortening, translating, tagging and linking. They all share the same glossary, the same tone of voice and the same data, configured per tenant.

Writing Assistant

Better texts without switching tools: the Writing Assistant shortens, sharpens and translates directly in the editor, in your brand’s tone of voice.

In development

Format variants & cut-outs

One image, every format: the crops capture the subject, and the original stays untouched.

In development

Tone profile & glossary

Your brand sounds the same everywhere: writing style and terminology apply when writing, translating and linking, in every language.

SEO assistant

Being found starts with writing: the hints appear where you are working, not in a report afterwards.

In development

Internal link suggestions

Pages that belong together find each other: the AI suggests links based on content, not keywords.

In development

Alt text & tagging

Accessible and findable from the start: alt text and keywords are ready as suggestions when you upload.

Multilingualism

You see where translations need to catch up

Whether it is the website, forms, notifications or the platform’s own interface: the German text is changed, the French version stays as it was, and nobody notices until someone gets in touch.

See what needs updating

When the German version changes, the stack shows every language version that needs updating and what has changed. This applies to page content as well as to the platform’s own texts.

The AI pre-fills missing or outdated translations, using your brand’s glossary and tone of voice. All languages stay at the same level, without anyone keeping lists.

AI discoverability

Being found, by AI too

Not everyone who visits your website is human: AI agents read, quote and recommend websites when they understand their content.

Readable for agents

Every website on the cloudtec Stack comes with what agents need to read it, and keeps it up to date automatically with every new piece of content.

This way, AI assistants understand what you offer instead of guessing it from the layout, and can cite your pages as a source.

Website of Beispiel AG with the page ‘Regrinding for carbide milling cutters’. Next to it, an AI assistant, asked who regrinds milling cutters in Switzerland, names Beispiel AG and cites the page as its source. The cloudtec Stack shows the MCP endpoint as connected.
From the blog

AI in action

Why we run AI ourselves, how it works in a client platform and how assistants work with your platform via MCP.

Company · 

We are building our own AI expertise

We run AI ourselves instead of just buying it in. Three reasons for the decision and three fields of application we are starting with.

Read the post: We are building our own AI expertise
See all posts

Where would AI make a difference for you?

Tell us what is costing you time. Together we check whether AI helps, and build the solution that fits your project, with or without AI.

Platform configurator Put your stack together step by step and get an initial assessment. Book a live demo Start a project