Euria AI at Infomaniak

Euria: What Potential New Users Should Know

Euria is an AI assistant developed by the Swiss technology company Infomaniak. Its main distinguishing features are privacy, digital sovereignty, Swiss hosting, and integration with Infomaniak’s broader productivity ecosystem.

Privacy and Digital Sovereignty

Euria is designed as an alternative to AI assistants operated by major US and Chinese technology companies. Infomaniak states that Euria is operated on its own infrastructure in Switzerland and that user data is not used to train the underlying AI models.

An ephemeral mode is available for conversations that users do not want retained. However, server-based privacy should not be confused with fully local AI: prompts are still processed on Infomaniak’s infrastructure. Users who require all AI processing to remain on their own computers or organisational servers would need a locally or self-hosted system instead.

Switzerland has strong data-protection laws and close legal alignment with Europe, but it is not a member of the European Union. Organisations with a strict requirement that data remain physically and legally within the EU should take this distinction into account.

What Euria Can Do

Euria is intended as a general-purpose AI assistant. Its capabilities include:

writing and rewriting;
translation;
summarisation;
document analysis;
audio transcription;
image analysis;
web-assisted research.

How Euria’s AI Works

Euria should be understood as an AI assistant platform, not necessarily as a single AI model developed and trained from scratch by Infomaniak.

An AI assistant consists of two broad layers:

The underlying AI model, which performs language generation and reasoning.
The surrounding assistant system, which provides the user interface, conversation management, document handling, web access, privacy controls, system instructions, integrations, and other tools.

Infomaniak can therefore deploy suitable open or open-weight AI models on its own infrastructure and build Euria around them. In such a system, a user’s prompt can be processed by a model running on Infomaniak’s servers rather than being sent to the company that originally developed the model.

This is important for digital sovereignty. Infomaniak does not need to spend billions developing every foundation model itself. It can benefit from advances in independently developed AI models while retaining control over the hosting infrastructure and user-facing service.

In principle, an assistant can also use different specialised models for different tasks, such as language generation, reasoning, image analysis, and speech transcription. The underlying models can potentially be replaced or upgraded without changing the Euria interface.

An Important Unanswered Question: Which Models Does Euria Actually Use?

For users evaluating Euria primarily on AI capability, this is an important issue.

Infomaniak publicly describes Euria as using advanced models selected for performance and efficiency, but the publicly available information examined for this comparison did not provide a sufficiently clear and authoritative list of the exact model variants used by Euria’s main assistant and their parameter counts.

References to model families such as Llama, Mistral, DeepSeek, or Qwen should not automatically be interpreted as confirmation that Euria currently uses all or any particular one of these models. Similarly, Infomaniak’s separate catalogue of AI models offered through its AI Services platform should not automatically be assumed to be identical to the models used internally by Euria.

This lack of clarity matters because a model family can contain versions of dramatically different sizes and capabilities. Knowing only that an assistant uses, for example, a particular model family does not reveal whether the deployed version has 7 billion, 32 billion, 70 billion, hundreds of billions, or a mixture-of-experts architecture with a different number of active parameters.

Why Parameter Count Matters—But Is Not Everything

Potential subscribers who care about maximum AI capability should investigate the exact underlying models before paying for Euria.

A 32-billion-parameter model, for example, can be highly capable and can outperform much larger older models on particular tasks. Modern training methods, data quality, architecture, reinforcement learning, and inference-time reasoning can all matter as much as raw parameter count.

Nevertheless, parameter count remains useful information when evaluating the approximate scale of a model. A relatively small model should not automatically be assumed to provide the same broad capability as a current frontier model simply because it performs well on selected benchmarks.

For mixture-of-experts models, two figures can be relevant:

total parameters in the complete model;
active parameters used to process each token.

Other important technical questions include:

What is the exact model name and version?
How many total parameters does it have?
For mixture-of-experts models, how many parameters are active per token?
Is the model quantised, and if so, to what degree?
What context window is actually available through Euria?
Does Euria route different tasks to different models?
Has Infomaniak fine-tuned or otherwise modified the underlying models?
Do free and paid users receive access to the same models?
What usage limits apply to the different subscription levels?

Until these questions are answered clearly, users should not assume that paying for Euria provides access to a frontier-scale AI model.

What Are Users Paying For?

Even if Euria uses a smaller open or open-weight model, the subscription can still have value. A hosted AI service provides considerably more than the raw model itself.

Potential value includes:

Swiss-hosted AI inference;
managed infrastructure;
privacy controls;
document and file handling;
web search;
audio and image processing;
integration with other Infomaniak services;
freedom from having to install, configure, update, and operate large AI models locally.

The key question is therefore whether a user is primarily paying for maximum model intelligence or for a private, sovereign, convenient AI service.

Integration with kSuite

Euria is particularly relevant to users of kSuite, Infomaniak’s productivity ecosystem. kSuite is broadly positioned as a Swiss alternative to Google Workspace and Microsoft 365.

Its ecosystem includes:

email;
calendar and contacts;
kDrive cloud storage;
collaborative office documents;
kChat for team communication;
kMeet for video meetings;
Euria AI functionality and integrations.

This can allow an individual or organisation to combine email, cloud storage, documents, communication, meetings, and AI within one Swiss-hosted ecosystem.

Advantages of Euria

Strong privacy orientation: Infomaniak states that user data is not used to train the underlying AI models.
Swiss hosting: AI processing can remain within infrastructure operated by a European-based Swiss provider.
Digital sovereignty: Euria reduces dependence on major US and Chinese AI platforms.
Integration: It can form part of a broader productivity environment through kSuite.
General-purpose capabilities: It supports writing, translation, summarisation, documents, transcription, images, and web-assisted tasks.
Managed service: Users do not need to own powerful GPUs or administer AI models themselves.
Environmental approach: Infomaniak emphasises renewable energy, energy efficiency, and reuse of heat generated by its data centres.

Disadvantages and Uncertainties

Unclear underlying model specifications: The exact models and parameter counts used by the main Euria assistant have not been sufficiently clear in the public information examined.
Potential capability gap: Depending on the models deployed, Euria can be less capable than leading proprietary frontier AI systems on difficult reasoning, coding, multimodal, or specialised tasks.
Parameter count is not enough: Even when model size is known, architecture, training, quantisation, context limits, and the surrounding assistant system also affect performance.
Not local processing: User data is processed on Infomaniak’s servers rather than entirely on the user’s own computer.
Switzerland is outside the EU: This can matter to organisations with strict EU-only data-residency requirements.
Smaller ecosystem: Euria and kSuite have a less extensive third-party ecosystem than ChatGPT, Google Workspace, or Microsoft 365.

Who Is Euria Most Interesting For?

Euria is particularly interesting for:

privacy-conscious individuals;
European organisations seeking greater digital sovereignty;
organisations wishing to reduce dependence on US Big Tech;
existing Infomaniak and kSuite users;
users who value managed, Swiss-hosted AI infrastructure more than guaranteed access to the largest available frontier model.

Bottom Line

Euria’s strongest proposition is currently privacy, sovereignty, Swiss infrastructure, and integration with Infomaniak’s ecosystem. Its attractiveness as a substitute for leading frontier AI assistants depends heavily on a question that potential paying users should investigate carefully: exactly which AI models are being provided, at what scale, and under what usage limits?

A user seeking a private, convenient, European-hosted AI assistant can find Euria attractive even if its underlying models are smaller than the largest frontier systems. A user whose primary requirement is maximum reasoning ability, however, should compare Euria’s actual performance and model specifications with competing services before purchasing a subscription.