Sovereign generative AI

Your sovereign AI,
deployed on your terms

Without depending on OpenAI or Microsoft. You keep control of your data: EU hosting, RGPD and AI Act compliance. Validated by a 3-week pilot, designed for public bodies and organisations.

GPU architecture · RAG · Mistral · OpenWebUI

413
Active users in production
200+
Concurrent users validated
100%
EU-sovereign hosting
The problem

Public AI is a data leak

Every prompt sent to a consumer AI is data leaving your security perimeter. Your users are already doing it. The only question: under what framework?

Total confidentiality

Your data never leaves a perimeter you control, hosted in the EU. No reuse to train third-party models.

RGPD & AI Act compliance

Architecture designed for regulation: EU data, structural minimisation, DPO documentation provided. You start compliant.

Technical independence

Auditable open-source components, no proprietary black box, reversible. You are never locked in.

Advanced RAG pipeline

From your documents to a sourced answer

Extraction (PDF, Office, OCR), vectorisation, re-ranking and inference, all on-premise, with citations of the passages used.

Your docsany format ExtractionPDF·Office·OCR Chunkingpassages Vectorsvector store Re-rankingBGE reranker LLMlocal inference Answersourced + cited
Clients

Already in production in local government

No slides, teams using our sovereign AI every day.

City of Annemasse & Annemasse Agglo : "MIA"

A sovereign AI built on open-source components (Mistral Small 3.2), in production for over 12 months. Label "Territoire Innovant · IA".

340
active users
1
single GPU
12+
months live

Calvados Department

The same sovereign foundation deployed at the scale of a department (~700,000 inhabitants): an AI assistant connected to the departmental SSO, EU hosting, open-source models.

73
active users
1
single GPU
SSO
connected

Interactive tools

Estimate the ROI of a sovereign AI, size the GPU (VRAM) for an LLM, check your AI Act risk level, and map automation potential by department. (Currently available in French.)

Technology stack

Auditable open-source components

No proprietary lock-in. Every brick is replaceable and auditable.

Open-source LLM

Mistral Small 3.2 24B Instruct in production, on a single GPU: sized to real need, no needless gigantism.

vLLM engine

High-performance inference, 200+ concurrent users validated in production.

Collaborative AI portal

Secure chat, knowledge bases, notes, group sharing, SSO login.

RAG + re-ranking

Multi-format extraction, embeddings, BGE reranker: sourced and relevant answers.

Observability

Grafana & Prometheus, email alerting, daily backups, automatic failover.

API first

OpenAI-compatible endpoints to integrate your existing business applications.

Test sovereign AI on your own documents

A 30-minute demo on a sample of your real documents, then a 3-week pilot (go/no-go milestone included). An expert replies within 48h.

Book a demo · 30 min
251 Route des Marais, 74250 Fillinges, France