On a warm June afternoon in Vienna, Matteo Rosoli stood in front of an audience at the Austrian Institute of Technology and made a pitch that's become increasingly familiar in European tech circles: We're building what Silicon Valley builds, but we're doing it here, under our rules, with our infrastructure.
The product is Eustella, an AI assistant that launched publicly on June 24, 2026, and bills itself as "the first fully sovereign European AI agent platform." Whether it's actually the first depends on how you define sovereign, and agent, and platform. What's less ambiguous is the timing. European regulators are tightening the screws on AI systems, enterprises are spooked by U.S. data exposure laws, and a clutch of European cloud providers and model makers are finally mature enough to support a consumer product that doesn't touch AWS, Azure, or Google Cloud.
Eustella's argument is deceptively straightforward: it does what the big American AI assistants do—chat, research, document generation, workflow automation—but runs entirely on European infrastructure, under European law, with zero dependency on U.S. tech giants. Rosoli shared the stage with executives from Mistral AI, IONOS Cloud, and Bitpanda, a lineup clearly designed to signal that the European stack is real and operational.
Whether that matters to most users is an open question. Whether it matters to regulators, procurement officers, and enterprises navigating GDPR and the incoming EU AI Act is considerably less so.
What You Get
Eustella is mobile-first, but it positions itself as more than a chatbot. The company describes it as "agentic"—meaning it doesn't just respond, it executes. Users can deploy pre-built "helpers" for tasks like deep research, news monitoring, or morning briefings. Or they can build custom agents and schedule them to run automatically. The platform reads, updates, and generates documents across PDF, Word, Excel, and PowerPoint. A slides-and-documents feature that launched June 8 added design template training and project management capabilities.
The use cases are practical, bordering on prosaic: travel planning, budget tracking, crypto portfolio monitoring, presentation generation. This isn't AGI—it's productivity software for prosumers and small teams. The interface is available on iOS, Android, and web. Pricing starts at free and scales to €5.99, €17.99, and €89.99 per month, all VAT-inclusive.
The feature set doesn't reinvent the category. Eustella's bet is that the infrastructure underneath it does.
The Sovereignty Play
Eustella's core pitch is about control, not capabilities. The company defines sovereignty as owning the full stack—models, compute, and data storage—without exposure to the U.S. CLOUD Act, which allows American authorities to compel U.S. companies to hand over data stored anywhere in the world. That includes EU data centers run by AWS, Azure, or Google Cloud.
"No AWS/Azure/GCP," the company states flatly on its sovereignty page. Instead, Eustella runs on IONOS in Germany for core servers, Scaleway in France for supplementary cloud workloads, Verda Cloud in Finland for GPU compute, and Mistral AI in France for specific OCR and transcription tasks via API. Linkup Technologies, also French, handles web search. Analytics are stored in Frankfurt via PostHog EU Cloud, though Eustella acknowledges residual CLOUD Act risk there given PostHog's U.S. parent entity.
There are exceptions, narrow but unavoidable. Apple's push notification service. Google's Firebase for Android authentication. RevenueCat for subscription management. These are ecosystem constraints—you can't ship an iOS or Android app without touching U.S. services at some point. Eustella notes these are covered by standard contractual clauses and data protection frameworks, a hedge that may or may not satisfy purists.
The company's multi-provider architecture is designed to avoid lock-in to major cloud platforms, though standard ecosystem constraints remain. It commits to transparency: its sub-processor registry, last updated June 23, names every entity that touches user data. For a startup pitching sovereignty to privacy-conscious users and enterprises, that level of disclosure isn't optional.
Models and Training
Under the hood, Eustella runs open-weight and open-source models: Qwen 3.5 for primary reasoning, Gemma 4 for speed and efficiency, an open-source large language model called GPT-120B OSS, various Mistral models, and Whisper for speech recognition. The reliance on Chinese-developed models like Qwen and DeepSeek raised eyebrows, which Eustella addressed in an April 10 blog post. The company's position: sovereignty is about who operates and hosts the models under EU law, not where the research originated.
It's a pragmatic stance, though perhaps not one that will satisfy everyone. The European AI ecosystem doesn't yet produce frontier-tier open models at the pace of Chinese labs or Meta. Eustella is working with what's available.
The company says it doesn't train on user data—a commitment repeated across its Trust Center and homepage. That's framed as both a privacy protection and a differentiation from Big Tech business models that monetize user content. Whether it's also a competitive disadvantage in the long run—no data flywheel, no continuous improvement loop—remains to be seen.
The Rollout
Eustella's public launch followed a measured rollout. A waitlist opened March 18. A closed beta started April 22 with over 2,000 signups. A web beta launched May 22 with 1,000 new spots. The document capabilities arrived June 8, weeks before the public launch. The pacing suggests a team that's cautious about scaling too fast—or perhaps one that's still ironing out infrastructure kinks.
The company is registered as AI Newsrooms Technology GmbH, based in Vienna, with a team LinkedIn pegs at 2-10 employees as of mid-June 2026. It raised a €750,000 pre-seed round in October 2025 led by Hermann Futter, with participation from Hansi Hansmann and the Hans(wo)men Group. That's seed-stage capital, enough to build a product and test a market, but not enough to compete head-to-head with OpenAI or Anthropic on feature velocity.
The launch event at AI Factory Austria featured representatives from Mistral AI, IONOS, and Bitpanda—strategic partnerships that lend credibility to Eustella's infrastructure claims and position it within a nascent European AI ecosystem. IONOS itself launched a "Momentum" AI ecosystem for SMBs and public sector clients in November 2025, explicitly pitching GDPR-compliant sovereign alternatives as a market opportunity.
First, or Not
Eustella's "first" claim is contested terrain, as these things tend to be. Atos launched something it called a "Sovereign AI Platform" in November 2024, but that was an enterprise, on-premises solution for large organizations—not a consumer-facing agent platform you can download on your phone. Meshton markets itself as "Europe's First Sovereign Confidential AI Agent Network," though it operates in a specialized crypto and confidential compute niche. The definition of "first" depends on how narrowly you draw the boundaries.
What's less debatable is the timing. The EU AI Act's deepfake disclosure obligations kick in August 2, 2026, according to Eustella's policy documents. European regulators are ramping up oversight. Enterprises are increasingly sensitive to where data lives and who can access it. Sovereignty, once a buzzword, is becoming procurement criteria.
Eustella is betting that a meaningful segment of users and businesses will care enough about architectural independence from U.S. tech giants to choose a European alternative, even if it doesn't match the feature depth or brand recognition of ChatGPT, Claude, or Gemini. The company's transparency around sub-processors and model choices, its critique of "Big Tech dependencies," its euro-denominated pricing—all of it points to a deliberate attempt to carve out a segment for whom trust and control outweigh cutting-edge capabilities.
Whether that segment is large enough to sustain a competitive AI platform is the open question. Eustella has infrastructure, partners, a product in market, and a clear point of view. Now it has to prove there's demand for sovereignty when the American alternatives are free, polished, and ubiquitous.
