Knodge.eu - Navigation & Developer
Welcome to Knodge.eu, the European Knowledge Hub.
The interview: "A PDF controls the machine"
Mareike (VBU) in conversation with Markus Schulte-Huermann about the future of adaptive AI systems.
**Mareike
*Hi Markus, the AI market is dominated by US giants such as OpenAI and Google. Why does the world - and European science in particular - need a platform like Knodge?
Markus
We do use Google's engine (Gemini), but we run it in an isolated container - similar to the way you can build a Ferrari engine in its own chassis. In Europe we drive a Ferrari differently and so we need a dedicated chassis.
The fundamental difference is: ChatGPT is a 'know-it-all' for the masses - great for poetry, bad for facts. Knodge, on the other hand, is a precision tool. We build an operating system that uses your own data to give valid answers instead of hallucinating the internet.
With knodge.eu, we are building a laboratory for specific, confidential knowledge. It's a kind of operating system or second brain - but simply more than just a chatbot.
Mareike
*In science, 'hallucination' (invented facts by AI) is a huge problem. How does Knodge ensure that research results remain valid?
Markus
We don't guess, we just search your content. We use the RAG principle (Retrieval Augmented Generation) for this. In this way, many small, evidence-based hubs can be created whose answer is a digital delivery bill made up of metadata, DOI, etc., so to speak. It's not the nice wording that counts here - it's the content and the source.
Mareike
*You talk about the 'Science Context Protocol' (SCP). What exactly does this mean for a modern laboratory?
Markus
This topic has been on my mind for a long time. The breakthrough for me came with a publication by Shanghai University on 30 December 2025. I immediately recognized the potential for robotics. SCP is the missing link. It is a standard protocol that not only reads data, but interprets it to trigger actions. Think of it like this: Previously, you needed an engineer to give the robot new parameters. With SCP, the robot reads the new PDF protocol itself and readjusts itself. The PDF controls the machine. This is the step from 'information' to 'instruction'.
At first this sounds very far-fetched, exaggerated and dangerous, but I think if you let it sink in a little, we will understand the necessity. We need to specify SCP globally for Europe.
It could also be used in the future to equip humanoid robots with knowledge or to define the scope of their tasks. Roughly speaking, a PDF controls the machine.
Mareike
*Data protection is sacred in the EU and at universities. Many institutions ban ChatGPT because data ends up on US servers. How do you solve this?
Markus
Oh yes - many people use ChatGPT without registering and then believe they are anonymous. But exactly the opposite is the case. Every request in ChatGPT is on the network and potentially visible. The IP address or the identification of the computer chip for anonymous users does the rest; identification is 99% guaranteed.
At Knodge.eu, everyone has to register with an e-mail address - that's all. After registration, there is no tracking or anything similar. A request to the AI runs via Knodge.eu as a portal request and is completely independent of the user. Google's AI is started individually for each portal. There is no direct connection between the user and the AI. All documents and content are stored on a bucket storage in the EU. The database with the control parameters is located on a completely different infrastructure, and the actual servers are also operated completely separately from the data. All data is stored anonymously and encrypted. We have no passwords that can be passed on and a genuine single sign-on for all users.
I don't know what else we could do to become even more secure. Google contractually guarantees us absolute 'private use'. Technically, this means: zero logging, zero training. Your sensitive research data will never be used to make the AI model smarter for others.
(Please note: PDF documents can optionally be sold explicitly via Knodge.eu; we offer these and can then be used to train specific models, but this is a separate process).
Mareike
*Research today is global and collaborative. How does Knodge help to break down knowledge silos between universities?
Markus
We are working on this as part of standardization in the SCP. We call it the new "Wrigge" function (beta). An example - the VUB can have one portal, the UKE Hamburg another. A higher-level hub can ask: "Are there contradictions between the results from Brussels and Hamburg?" without having to copy the raw data. This enables meta-analyses in real time across institutional boundaries. In this way, we turn passive paper knowledge into active process control (SCP) and thus provide an answer to researchers' data concerns. Private AI, not training. We want scientists to spend less time searching and formatting and more time actually discovering. Knodge is the tool for this.
Mareike
*One last technical question: Many institutions have extremely strict compliance rules and are not allowed to use external clouds at all. Are these then excluded from Knodge?
Markus
No, on the contrary. That's the beauty of our architecture: Knodge is completely containerized.
That means:
We do offer our secure European cloud, but we can also install a Knodge instance "on-premise", i.e. directly in the university's or company's data center. You then operate the system completely independently. And thanks to the "Wrigge" function, you can still securely connect this local island to other partners if you wish. Whether in the cloud or in your own basement - the technology remains the same.
Mareike
*Thank you for the interview, Markus.