Swiss Sovereign AI

AI for the data
you can't send away.

We fine-tune Apertus — Switzerland's open LLM from EPFL, ETH Zurich and CSCS — on your documents and run it inside your infrastructure: your datacentre, your Swiss cloud tenancy, or fully air-gapped. Open weights your auditors can inspect. No US API in the loop. You own the model.

Book a call How it works

Built on Apertus by EPFL, ETH Zurich & CSCS — open weights, trained on the Alps supercomputer in Lugano, already running in production in the Swiss public sector.

Third-party US LLM API

Your jurisdiction

01

Your data

Documents, records, systems — never leaves
02

Fine-tuned Apertus

Swiss open model, trained on your corpus
03

Agent orchestration

Agents · RAG · tools · guardrails
04

Your product

Copilots & workflows — yours to run

On-prem · Swiss cloud · Confidential compute — zero data egress

Compliant by design Swiss FADP / nFADP EU GDPR EU AI Act documentation Swiss jurisdiction No data egress Open weights

The problem

Frontier APIs weren't built for regulated teams.

For banks, hospitals, insurers and public bodies, external AI isn't forbidden — it's slow and residual-risk-heavy: months of legal review, DPAs, sub-processor chains, and at the end a model you still don't control. Most projects get stuck at the pilot stage — impressive demos that never clear the review.

Data leaves your control

Prompts and documents are sent to third-party servers in another jurisdiction.

No transparency

Closed weights mean no auditability — a non-starter for regulators.

Vendor lock-in

Pricing, availability and model behaviour can change without notice.

Stuck at PoC

Demos don't survive a data-protection or security review.

sysf.io removes the blocker: a model you own, running where your data already lives.

And this isn't theoretical — the Canton of Ticino runs a fine-tuned Apertus-8B translation service in production on its own infrastructure, reaching 94% accuracy after fine-tuning on cantonal data. The same playbook applies to your workflows.

Why sysf.io

Six reasons this architecture clears compliance review.

Swiss by foundation

Built on Apertus, Switzerland's sovereign open LLM — open weights and documented training data, not a black box.

Compliant by design

Built for Swiss FADP/nFADP and EU GDPR, with the documentation your EU AI Act obligations demand. Transparency and auditability come from the architecture.

Sovereign & private

Your data never leaves your jurisdiction. On-prem, Swiss cloud or confidential compute — no US API dependency.

Customized on your data

Domain fine-tuning plus RAG and agentic tools on your own corpus, systems and terminology.

End-to-end delivery

Data → fine-tune → evaluate → deploy → monitor → maintain. A maintained production system, not a proof-of-concept.

Multilingual, incl. Swiss

Native German, French, Italian and Romansh — including Swiss German — from a model built for Switzerland, not translated to it.

Ownership

Built to be fired.

Everything we build is yours. That's the point of open weights: no rented model, no lock-in, no hostage negotiation at renewal time.

You own the weights

The fine-tuned model is yours — run it with us, without us, or with any team you choose.

You hold the runbooks

Data pipelines, evaluation harness and operational runbooks are deliverables, not secrets.

Handover is a deliverable

Every engagement includes a documented handover. Your exit strategy is part of the scope, not a negotiation.

The platform

From your data to a product in production.

A repeatable, end-to-end pipeline — every step runs inside your jurisdiction.

Step 1

Scope & comply

Use-case discovery and compliance scoping (FADP, GDPR, EU AI Act).
Step 2

Fine-tune on Apertus

Private, in-jurisdiction data prep and domain fine-tuning.
Step 3

Orchestrate

Agents, RAG and tools wired to your systems, with guardrails.
Step 4

Deploy & operate

Sovereign deployment, monitoring, evaluation and iteration with support agreements.

Explore the platform

Industries

Built for Switzerland's regulated sectors.

Banking & Finance →

Advisory copilots, KYC/AML document processing, regulatory Q&A — on-prem, zero data egress.

Healthcare & Pharma →

Clinical documentation and research assistants — patient data stays in Switzerland.

Legal & Compliance →

Multilingual contract analysis and legal research (DE/FR/IT), privilege-safe.

Government & Public →

Sovereign, multilingual citizen services on national infrastructure.

Insurance →

Claims automation and policy Q&A copilots grounded in your documents.

Manufacturing →

On-prem knowledge and maintenance assistants for the shop floor.

Engagement

Start small. Prove value. Scale to air-gapped.

Every tier is sovereign — your data stays in your jurisdiction; tiers differ in deployment depth. Pricing is scoped per engagement — see the model.

Pilot

Prove value on one use case

Fine-tuning and retrieval on your data, an evaluation report against agreed metrics, and a production roadmap.

Production

The pilot, hardened for daily use

Agents, RAG and guardrails wired into your systems, with deployment, monitoring and support.

Air-gapped

Regulated / most sensitive

Fully on-prem or confidential compute, dedicated operations, and complete compliance documentation.

See the engagement model

FAQ

Answers for evaluators.

What is sysf.io?

sysf.io is a Swiss sovereign-AI engineering studio. We fine-tune Switzerland's open Apertus model on your data and ship it as a production system — retrieval, agents and guardrails wired to your systems, deployed in Swiss or EU jurisdiction, or fully on-premise. The result is a compliant, private, customized AI system you own — not a demo.

Which model do you use — is it always Apertus?

Apertus is our foundation of choice — Switzerland's fully open, multilingual LLM from EPFL, ETH Zurich and CSCS, with open weights and documented training data, which is exactly what auditors and regulators want to see. But we're not dogmatic — we test open-weight models against an evaluation set built from your documents and deploy whichever wins. Apertus, Llama or Mistral — the evaluation decides, not the logo. All of them run fully inside your jurisdiction with weights you can inspect and own.

Does our data ever leave our environment?

For Production and Air-gapped engagements — no. Training and inference run inside your tenancy or your datacentre, with no third-party LLM APIs and no data egress. Pilots run either in your environment or in a dedicated single-tenant Swiss cloud environment under a data-processing agreement, with all data and derived weights deleted or handed over to you at project end — your choice at scoping.

How is this "compliant by design"?

Compliance is built into the architecture rather than bolted on. Open weights give you transparency and auditability; in-jurisdiction deployment addresses the Swiss Federal Act on Data Protection (FADP/nFADP) and EU GDPR; and documented data lineage plus guardrails support your EU AI Act obligations. We deliver the compliance and audit documentation alongside the model.

What does "end-to-end" mean here?

We take a use case from data preparation and fine-tuning through evaluation, guardrails, deployment, monitoring and ongoing iteration. You get a maintained production system with agent orchestration, retrieval (RAG) and tools wired to your systems — plus a support and maintenance agreement — not a proof-of-concept that stalls.

How do we start?

Engagements start with a fixed-scope Pilot on a single high-value use case — prove the numbers on your own data, then decide about production with an evaluation report in hand. Where needed, we take the system fully on-prem. Book a call and we'll scope it with you.

Ready to own your AI?

Book a scoping call and we'll map your first sovereign use case — on your data, in your jurisdiction.

Book a call See how it works

AI for the datayou can't send away.