Independent AI Consulting & Architecture · Zurich
Your data is confidential.
Your AI must be too.
Souverana makes AI a reality in your organisation: an AI strategy that is sovereign, confidential and production-ready, and support with implementation until AI works inside your processes and products. The work only really begins after the concept: Souverana builds with you and stays.
Mandates from one-person firms to Fortune 500 corporations.
Three things often mistaken for an AI strategy.
Copilot switched on
Turning on Microsoft Copilot and hoping the rest takes care of itself is not an AI strategy.
Licences rolled out
Giving your developers GitHub Copilot and waiting for productivity is not an AI strategy.
A title awarded
Naming someone "AI Engineer" and having them wire up a RAG prototype is not an AI strategy.
If you recognise yourself here, you are in good company. An AI strategy starts somewhere else: with the question of where AI creates value in your business, which data it needs for that, and how it reaches production safely.
Sovereignty means having control, being independent and being able to act on your own terms.
Sovereign AI consists of four layers.
AI that never gives your data away is built on four layers: from strategy to the integration into your business applications. Good, open options exist for every layer today. Souverana selects them and puts them together.
- 1
Strategy
Where do you stand today, where does AI create value in your business, and in what order do you proceed? Every engagement starts here.
- 2
Infrastructure
Language model and hosting, vector database, document store, data and knowledge pipelines: the foundation everything runs on.
- 3
Platform
The tools your team works with every day, wired into one whole: Open WebUI, OpenCode and whatever else fits the way you work.
- 4
Integration
Your own agents, RAG and MCP connections to your systems: this is where AI merges with your business.
Around the four layers sits the culture. Organisations that make progress with AI experiment early and get used to change. Souverana brings that attitude to every engagement.
Sovereign AI is ready for real work today. The building blocks exist.
Public administrations, hospitals and law firms can run AI productively on confidential data today, in Switzerland and under their own control. The technology is not the bottleneck. What it takes is someone who assembles the building blocks so they hold for your data.
Shadow AI: The demand is there, the official path is not
Your employees already use ChatGPT, including with case files and patient data. The task is real and the tool works. The same task can now be solved on a path where your data stays under your control. The first step is to make that path official.
The building blocks are here
Open models like Apertus, hosting in Switzerland or on your own premises, mature tooling for RAG and MCP. Souverana assembles them into a stack that holds up even under official and professional secrecy.
The legal frame is part of the design from the start
The revDSG governs how personal data is handled, and the EU AI Act applies in addition depending on the case. Documentation and enabling your staff are planned in from the start. Souverana carries these requirements along as construction details, so the architecture fulfils them from the start.
Independent of platform and vendor
Hyperscalers, platform vendors and large consultancies end up recommending their own solution. Souverana sells neither a platform nor a model. The recommendation therefore follows your situation.
The AI Sovereignty Check.
13 questions, two minutes, no sign-up. You get your requirements profile, your maturity level and a concrete assessment of what is already possible with your data today. Plus your next three steps.
The evaluation runs entirely in your browser.
How an engagement works.
You start small and with a clear frame: every step produces a result the next one builds on.
-
01
AI Readiness Assessment
Where you stand today, which data already flows into external AI services, and which duties the revDSG and the AI Act place on you. It ends with a prioritised path forward.
-
02
Architecture & Solution Selection
The sovereign target architecture across all four layers, with independent selection of model, hosting and platform.
-
03
Implementation & Integration
Building the solution and coordinating the hosting and operations partners.
-
04
Ongoing Support
Souverana stays on board as your AI architect: operations, rollout, further development.
References.
Souverana is the AI consultancy of Joel Barmettler and works for organisations of every size, from one-person firms to international corporations. Three current mandates, across the four layers:
Integration: MCP servers for a Fortune 500 corporation
An international Fortune 500 company (name confidential) is opening its own software to AI through MCP servers. Souverana helps build an environment in which AI can access a large body of data, reason over it and act on it.
Platform & infrastructure: Swiss AI Hub
The Swiss AI Hub was originally built by Joel Barmettler in his role as AI Architect at bbv Software Services AG. Today Souverana shares responsibility for the architecture of the open-source project: a complete AI stack that solves exactly this infrastructure and platform problem. Deployable with a single command, air-gapped on a single GPU or in any Swiss cloud.
Strategy: AI strategy for a Swiss SME
Souverana supports a Swiss SME (name confidential) with its internal AI strategy, focused on AI in software engineering and a sovereign AI transformation.
Frequently asked questions.
- May we use AI with patient data, case files or client data?
- Yes, provided the processing stays under your control. The revised Swiss Data Protection Act (revDSG) does not prohibit AI; it demands control, transparency and proportionality. With processing in Switzerland or on-premise, clear permission concepts and traceable sources, AI can be justified even under official and professional secrecy.
- What is the privacy-compliant alternative to ChatGPT and Microsoft Copilot?
- A stack of open building blocks: a language model such as the Swiss Apertus, hosted in Switzerland, with a platform and an integration layer you control. For most business tasks such as knowledge search, document analysis or assistance, this reaches production-grade quality.
- Does the EU AI Act oblige us to do anything?
- In many cases yes, even as a Swiss organisation. Depending on the use, documentation, transparency and competence duties apply; employees must receive AI training. The AI readiness assessment clarifies exactly how you are affected.
- Does Souverana build as well, or only advise?
- Both. Souverana builds the architecture and the integration layer (agents, RAG, MCP connections) itself. For hosting and operations Souverana brings proven Swiss partners on board and coordinates them. You have one contact who stays through to operations.
- What does adopting AI cost?
- Transparent from the first step: the AI readiness assessment from CHF 3’500, a target architecture from CHF 8’000, implementation at a day rate of CHF 2’100, ongoing support as a Fractional AI Lead from CHF 8’400 per month, all excl. VAT. The intro call and the AI Sovereignty Check are free.
Put AI to work on your real data.
A free 30-minute intro call clarifies what is possible first in your organisation and what the next concrete step looks like.