Sovereign AI in Switzerland — on-premise deployments

Your data stays in Switzerland, your models run in-house. Open-weights LLMs on on-premise GPUs, native nLPD compliance, zero dependency on foreign clouds. Concrete Swiss sovereign AI.

01 · Typical use cases

Four pillars of sovereignty.
One shared architecture.

Swiss sovereign AI is not a marketing claim: it is a precise technical architecture. No data transits to servers outside your perimeter. Models run on your infrastructure and your legal obligations (nLPD, FINMA, professional secrecy) are satisfied structurally, not contractually.

Open-weight models from DeepSeek, Mistral and Qwen can run on infrastructure you control. Cost and quality depend on the model, hardware and workload.

SWISS PERIMETER LLM Denied
SVC.001 · PERIMETER

Data that doesn't leave

Model and data hosted on your infrastructure — your premises or a datacenter under direct contract. No transit to foreign servers, no exposure to the Cloud Act.

AIRGAPON-PREMZERO-EXTERNAL
COMPARE MODELSAnswer qualityResponse timeCost per task
SVC.002 · OPEN-WEIGHTS

Performant open-weights models

DeepSeek-V4.1-Flash, Mistral Large 4 and Qwen3.8-27B are examples of open-weight models. We compare quality, latency and cost on your tasks.

DEEPSEEKMISTRALQWEN
AUDIT TRAILAccess controlLocal hostingAudit logs
SVC.003 · COMPLIANCE

Structural compliance

nLPD and professional secrecy satisfied by the architecture — not just by contracts. Processing register, access control and audit logs delivered.

nLPDAIRGAPPRO-SECRECY
ROUTER Private · local Public · cloud
SVC.004 · HYBRID

Hybrid sovereign · request router

Sensitive requests go to the local LLM, generic requests to a public API. A router based on content classification automates the dispatch.

ROUTERDUAL-PATHFALLBACK
02 · Our approach

Scoping, benchmark, sovereign deployment.
Native compliance, not contractual.

Before any decision, we benchmark several open-weights models on your real data — not on generic leaderboards. Deployment happens on your infrastructure (your premises or a datacenter under direct contract), with configured GPUs, inference and documented compliance.

Step 01

Sovereign scoping

Inventory of sensitive data, nLPD/FINMA constraints, targeted on-premise perimeter. The sovereignty need is clarified before any technical choice.

Step 02

Benchmark on your data

Several open-weights models evaluated on your real corpus — quality, latency, GPU cost. The recommendation rests on concrete metrics, not on a leaderboard.

Step 03

Deployment & governance

vLLM inference on your GPUs, secrets in Vault, LDAP access, audit logs. nLPD documentation (processing register, access rights) delivered with it.

03 · Models & infrastructure

Our open-weights models
and the on-premise infra.

// open-weights models
01
DeepSeek · text and images
DeepSeek-V4.1-Flash

Multimodal model with weights published under the MIT licence. Its deployment requires suitable infrastructure.

02
Mistral AI · European
Mistral Large 4

Open-weight multimodal model under the Mistral Ecosystem Open Weight licence.

03
Alibaba · multilingual
Qwen3.8-27B

Open-weight variant for text and images. Test it on your documents and hardware.

// sovereign infrastructure
04
On-premise inference
vLLM · NVIDIA GPU

Continuous batching, KV cache, tensor parallelism on A10G / A100 / H100 in your datacenter.

05
Data & access
Qdrant · Vault · LDAP

Local vector store, secrets in Vault, LDAP access control integrated with your directory.

The hybrid sovereign architecture — sensitive requests to the local LLM, generic requests to a public API — is today our default recommendation. It simultaneously satisfies nLPD requirements and professional secrecy obligations (banking, medical, legal).

04 · FAQ

Frequently asked questions.

05 · Go further

Related services.

Reply within 24 business hours

Got a use case in mind?
Let's talk.

Costed audit, measurable prototype, sovereign deployment. No sales middleman — you speak directly to a member of the technical team.

For companies based in Lausanne (Vaud), Geneva, Neuchâtel, Fribourg, Jura and Valais. Learn more about our AI agency.