nenna.ai develops AI privacy platform on Exoscale's European cloud infrastructure

nenna.ai was founded in 2023 in Berlin by a four-person team with particular expertise in natural language processing and machine learning. The AI privacy startup addresses a challenge that is putting increasing pressure on companies: how to use the capabilities of large language models without exposing sensitive data. The answer is an AI privacy platform that masks personal and confidential information in real time before it reaches a model, then translates the model's response back into fully readable form. The model sees only anonymised data points; users receive their results in plain text. The entire platform runs on Exoscale's European cloud infrastructure, with server locations in Germany and no data transfer to third countries.

The nenna.ai offering consists of four products: 

• NENNA Chat gives companies a secure, internal AI interface where employees can use any model they choose, including exclusively European providers if preferred. 
• NENNA Watch protects the use of external AI tools via a browser extension, preventing sensitive data from flowing out uncontrolled and preventing shadow AI use altogether. 
• NENNA Shield allows companies to integrate the privacy functions into their own applications via an API. For the legal sector, nenna.ai has developed. 
• NENNA Legal, a specialised solution that meets the requirements of professional legal confidentiality.

The company has already won the Legal Tech Award at the German Legal Tech Hub as the best legal tech startup in Germany in 2025 and the ATHENE Award from the Fraunhofer Institute as one of the three leading cybersecurity startups in the DACH region.

The Challenge

Shadow AI is no longer a fringe phenomenon in companies. Employees upload documents to public AI chats, ask questions about client projects or have contracts summarised, and in doing so, confidential information ends up on servers that the company neither knows nor controls. Dr Max Flöttmann, CTO and co-founder of nenna.ai, identifies the core dilemma: “Tonnes of company data end up in free public accounts because it helps employees do their jobs. And if you don't simultaneously regulate AI use while enabling efficient working with AI, you lose your data.”

Banning AI outright drives uncontrolled use further underground. nenna.ai's answer is therefore not restriction but managed access. Companies receive a platform that makes AI productively accessible without surrendering control over their data. To deliver on this promise reliably to enterprise customers, nenna.ai needs an infrastructure that meets the same requirements.

Finding the Right Infrastructure

When the team began building the platform in early 2024, Hetzner provided parts of the infrastructure. The offering was sufficient for initial development steps but quickly reached its limits. nenna.ai had to build and operate the entire cluster infrastructure on its own, databases ran on dedicated servers, and scalable managed services were absent. For a small team that wanted to direct all its engineering capacity into product development, this was not a sustainable foundation in the long run.

A US hyperscaler was ruled out as an alternative from the outset. A company that promises its customers to protect their data on European infrastructure cannot entrust that data to a US company. From earlier professional roles, the team was also familiar with what Flöttmann calls the “golden cage” effect. A provider starts with one convenient service, then adds another, then another, and before long the system is so deeply embedded in a single provider's ecosystem that switching has become practically impossible. Open standards and portability were therefore non-negotiable for nenna.ai from the very beginning.

In a comparison of several providers, Exoscale, the European cloud infrastructure provider from A1 Digital, stood out for several reasons. Unlike many competitors, the development team could immediately test services and processes without first having to sit through multiple sales calls. When standard support could not resolve an issue, engineers were directly reachable and could solve problems without delay. Because Exoscale is built on open standards, nenna.ai is not tied to a single provider. The company can bring in or switch to another provider at any time without having to build up specialised knowledge of a proprietary platform. Flöttmann summarises the decision: “From my perspective, Exoscale works like a hyperscaler, at least for a company like ours. And I can use it just as easily as I was previously used to using AWS. The fact that Exoscale provides European infrastructure is not a minor detail for a company like nenna.ai — it is the decisive difference.”

The nenna.ai Platform in Action

nenna.ai today uses several Exoscale services to operate its platform. At the core is a managed container infrastructure on which the entire application logic runs, complemented by cloud-based object storage for data management and managed databases.

Real-time masking places particular demands on latency and computing capacity. As a user types, the system must immediately detect whether personal or confidential data is about to leave the company. For this, nenna.ai relies on specialised, smaller AI models that respond significantly faster than their larger counterparts.

Exoscale also made it possible to extend the functionality of NENNA Chat. The solution now has a cloud instance on Exoscale that allows the model to work directly with uploaded files, from creating documents and analysing spreadsheets to populating company templates, while the masking layer remains continuously active.

Results

Since launching on Exoscale in early 2025, nenna.ai has grown its infrastructure at least twentyfold. What began with a handful of instances is now a complete, scaling cloud infrastructure that has grown alongside the demands of the platform. Customers including Duisport, the largest inland port in Europe, the Mazak media group and IWB use the platform, in some cases already company-wide.

The engineering capacity at nenna.ai that would previously have gone into manual infrastructure operations now flows entirely into product development. For a small team, that matters more than the equally significant cost savings. Flöttmann puts this down to the quality of the working relationship from day one: “Support proved to be very helpful from the very beginning. It was always a very pleasant collaboration, between genuine partners.”

Outlook

nenna.ai is working on extending masking to external data sources and is simultaneously pursuing further certifications. Following its existing ISO 27001 certification, the company will begin the ISO 42001 process in autumn 2026, the AI-specific standard for model transparency. On the infrastructure side, nenna.ai and Exoscale are continuing to develop their collaboration to keep pace with the platform's growth and expanding scope.