mentro logo
Hosting & security

Hosted in Switzerland.

From the servers to AI inference: all data created in mentro stays in Switzerland. Here we show transparently where your data lives and who we work with.

Our infrastructure

Where your data lives

Compute & operations

cloudscale.ch

The mentro platform runs on cloud servers by cloudscale.ch with data centres exclusively in Switzerland. cloudscale.ch is ISO 27001 certified.

Database

peaknetworks, Gais

We operate our database ourselves, on peaknetworks servers in Gais, in an ISO 27001-certified Swiss data centre. No US cloud provider as a subprocessor.

AI

Infomaniak

The AI for automatic competency mapping runs at Infomaniak in Switzerland. Infomaniak is ISO 27001 certified.

Security in detail

Data stays in Switzerland

All data created in mentro sits on servers in Switzerland and never leaves the country. We use no US providers for the platform. Outside the platform, our own correspondence runs on Microsoft 365, for example emails from the contact form.

Encrypted in transit

Everything travelling between app and servers is encrypted. Nobody can read along on the way.

Every company stays separate

Every company sees only its own data. The separation sits in the database itself, not just in the app.

Privacy-first AI

Before every AI analysis, names as well as health and family details are pseudonymised. The AI only maps, it does not grade.

Read the privacy policy

Swiss Made Software + AI

What AI means at mentro

mentro carries the swiss made software + AI label. Here is what that means in practice, without the marketing: which models run, where they run and where we deliberately draw a line.

Operating model
mentro is a SaaS offering. The platform runs at cloudscale.ch, the database at peaknetworks in Gais, the AI inference at Infomaniak. All three operate their data centres in Switzerland.
Depth of integration
We operate no foundation model of our own. Third-party models sit inside our own processing chain: sanitise the input, narrow the candidates with a vector search across the performance objectives of your training plan, then the language model decides. If it fails, the vector search result stands.
Technologies used
A large language model for mapping to the training plan and for selecting the reflection questions. Alongside it, multilingual text embeddings for the semantic pre-selection.
Model origin and type
Open third-party models, operated in Switzerland: Mistral Small 3.2 as the language model, BGE Multilingual as the embedding model. No foundation models of our own, no closed models from US providers.
Adaptation and fine-tuning
We train no models and do no fine-tuning. Adaptation happens through system prompts, through curated keyword lists on the performance objectives and through the vector search in front of the model. Company data feeds into no model training.

Before every AI call, names as well as health and family details are pseudonymised. The AI maps and suggests, it does not grade. The education report is written and signed by the trainer.

Our hosting partner

Swiss cloud by cloudscale.ch

To run the platform we rely on cloudscale.ch: Swiss cloud infrastructure, transparent and reliable. Exactly the foundation sensitive education data deserves.

Visit cloudscale.ch

Questions about data and security

All platform data stays in Switzerland. The servers are at cloudscale.ch, we run the database at peaknetworks in an ISO 27001-certified Swiss data centre, and the AI runs at Infomaniak. We use no US providers for the platform. Outside the platform, our own correspondence runs on Microsoft 365, for example emails from the contact form; the privacy policy sets this out. Transmission is encrypted (TLS), and every company sees only its own data.

You can export your data at any time as JSON (education reports also as PDF). After the contract ends we keep it available for export for 90 days and then delete it, within the scope of data processing and the legal retention obligations of the training company.

Not generated, but written by hand and carefully selected. mentro maintains a pool of over 80 pedagogically developed reflection questions in five stages: from simple description through analysis to comparing several perspectives. Each period, an AI picks the five that fit, matched to the learning year, the competency and the learner's actual activities. Every reflection is guided, but never schematic.

App, website and the learning journal questions are available in German, English, French and Italian. You can switch the language at any time in the menu.

All questions and answers