AI Models

eustella is powered by open-source / open-weight or European AI models, hosted entirely on European infrastructure. We are not dependent on any proprietary model, vendor or API, since we host the models ourselves. No AI model will process data outside of Europe and there will never be a single vendor lock-in.

Open models. European control.

The real question is not where a model was trained — it is who controls it. A closed model accessed through a US API sends every request outside European territory. An open model, hosted in Europe, stays under European control.

eustella selects the strongest open-source / open-weight models available — regardless of origin — and runs them on EU-based servers. Open-source / open-weight models can be audited, fine-tuned, and taken offline entirely. Closed API models cannot. That difference is what makes genuine data sovereignty possible.

This is the pragmatic path: it is not always about where a model was born — it is about who runs it, who audits it, and whose laws govern it. The same logic applies in the energy sector, where Europe pursues independence with the help of solar panels manufactured elsewhere. Sovereignty is about control, not origin.

That said, eustella dreams of a future where Europe produces its own world-class foundation models — trained on European data, shaped by European values, and competitive at the global frontier. Building that future takes time, funding, and a generation of European AI talent. And who knows — the next great European foundation model might come from closer to home than you think.

Learn more about eustella's approach to sovereignty

The models behind eustella

eustella always runs several models at the same time. Each request is routed to the model best suited to it — reasoning depth, speed, voice, or long context — so no single model can ever become a dependency.

Gemma 4

Fast, efficient language model

eustella uses Gemma 4 for tasks that require speed and efficiency — quick lookups, summarisation, and lightweight interactions where latency matters more than depth. Developed by Google DeepMind, hosted by eustella in the EU.

Voxtral

Speech understanding

eustella uses Voxtral for voice input and audio understanding. Voxtral is an open-source / open-weight speech model family released under the Apache 2.0 licence, handling both transcription and direct questions about audio across the major European languages. Developed by Mistral in Paris, hosted by eustella in the EU.

Whisper

Speech recognition

eustella uses the open-source / open-weight version of Whisper for voice input and transcription. Whisper supports all eight languages eustella serves and runs entirely on European infrastructure — your voice data never leaves the EU.

Qwen Model Families

Flagship reasoning models

eustella uses the Qwen model families for its most demanding reasoning work. Qwen models are among the most capable open-source / open-weight models available, with strong multilingual performance across all major European languages. Developed by Alibaba Cloud, hosted by eustella in the EU.

Mistral Models

European language models

eustella integrates models from Mistral, the leading European AI lab based in Paris. Mistral models are particularly strong in French, German, Spanish, and other European languages — a natural fit for a European AI assistant.

GPT-120B OSS

Open-source / open-weight large language model

eustella uses GPT-120B OSS for complex, long-context tasks that benefit from a large parameter count. A fully open-source / open-weight model with no API dependency and no data shared externally.

The open-source / open-weight AI landscape evolves fast. eustella continuously evaluates new models as they are released and will adopt stronger alternatives whenever they become available. Some of the models listed here may be replaced, and new ones may be added. eustella will always publish the current model stack on this page so you know exactly what powers your AI assistant.

Why open models matter

No vendor lock-in

eustella can switch, combine, or replace any model at any time. If a better model appears tomorrow, eustella can adopt it without rewriting the product or renegotiating a contract.

Full auditability

Open-source / open-weight models can be inspected for bias, censorship behaviour, and safety. Closed models require trust. Open models allow verification.

European hosting

Every model eustella runs is hosted on EU-based infrastructure. No API calls to US or Chinese servers. No data leaving European jurisdiction.

Cost independence

eustella does not pay per-token fees to Big Tech API providers. Self-hosted models mean predictable costs and no dependence on external pricing decisions.

Using AI safely

AI models are powerful tools, but they have real limitations. eustella believes you should understand what AI can and cannot do — including hallucinations, sycophancy, and when to seek human expertise instead.

Read our AI safety guide

Get started with eustella

Create your free account — eustella works everywhere, on web, iOS, and Android.