Sovereign managed inference
We operate for you the building blocks we produce, on platforms subject neither to the Cloud Act nor to any other extraterritorial law. Bespoke, at your level of requirement, with the ability to take everything back.
What we operate
The building blocks we produce, run for you, to the standard you would apply yourself.
- Open-RAG.ai
Conversational assistants on your corpora
Search and sourced answers over your internal documents, with every answer traceable back to the original passage.
- LinTO.ai
Transcription and speech processing
Transcription, translation and meeting summaries, in real time or after the fact, on your streams and your archives.
- Luciole
Hosted language models
Luciole at 1, 8 or 23 billion parameters, and a selection of third-party open models, served on our sovereign platforms or on yours.
- Twake AI
Augmented collaboration
Messaging, video meetings and documents augmented by AI, without your internal exchanges leaving the agreed perimeter.
- AI Augmented Teams
Business agents on an interoperable harness
Agentic loops specified with you, on a modular harness where each piece can be replaced independently.
Four deployment modes
The right mode depends on what can leave your premises, and on what you must be able to prove.
Your machines, your walls, our operations. We work under your access rules, with named and identified staff.
For : Entities that cannot let anything leave their perimeter.
Dedicated instances at a European host not subject to extraterritorial laws, with physical or logical isolation depending on the level required.
For : Organisations that want isolation without running a machine room.
Shared resources and strict partitioning, for a markedly lower entry cost and a faster time to service.
For : Organisations whose sensitivity is real but whose volume is modest.
Sensitive processing on your side, elastic load on a sovereign platform. The trade-off is made use case by use case, defined with you and written down.
For : Organisations whose uses do not all carry the same level of sensitivity.
What we implement on the sovereign clouds
On the hosted modes, we implement the measures required for compliance with the AI Act, the SecNumCloud 3.2 framework, the NIS 2 directive and the GDPR. These requirements partly overlap, but they do not substitute for one another: each is addressed for what it asks, technical documentation, risk management, systems security, data protection.
GPUs can optionally be hosted on SecNumCloud-qualified infrastructure, for organisations whose level of requirement justifies it.
What we guarantee
Every point below is written into the contract, not only onto this page.
Location of data and processing
Data, processing, backups and logs all remain within the agreed perimeter. Backups and logs are the two places where location is most often lost, so they are handled explicitly.
No extraterritorial exposure
The hosting falls neither under the Cloud Act nor under any other extraterritorial law. The commitment also covers second-tier subcontractors, whose list is shared with you and whose every change is notified to you.
Identified operations staff
You know who operates your service, from which country and under which legal regime. Access is named, traced and reviewed periodically.
Encryption and key custody
Encryption in transit and at rest. Key custody is settled explicitly during scoping: you hold them, we hold them, or they are shared. That choice determines what we can do, and what we cannot.
Service level commitment
Availability, restoration time, maintenance windows and penalties, defined per deployment mode and written into the contract.
Reversibility
Export format, timeline, exit support. And above all: because the building blocks are open, leaving is technically possible without us. Our support makes it faster, it is not what makes it possible.
Availability figures, restoration times and penalties are contractual and depend on the deployment mode chosen. They are established during scoping.
AI Act compliance
A managed service does not make your organisation compliant. It supplies part of what compliance needs, and it is worth knowing which part.
What we supply
- Technical documentation of the systems we deploy and operate for you.
- Traceability of the training chain for the models we produce, datasets and method included.
- Logging of processing, within the agreed perimeter and for the agreed duration.
- Our contribution to risk assessment, on the technical side of the system.
- Explicit designation of the roles defined by the regulation, provider and deployer, for each component.
What remains yours
- Classifying your systems by risk level, which depends on the use you make of them.
- Assessing risks to the people concerned, which belongs to your business and not to ours.
- Informing users and, where the regulation requires it, providing human oversight.
- Internal governance: who decides, who checks, who answers.
- The compliance of third-party models you ask us to host, when their training documentation is not public.
We do not sell turnkey compliance: it does not exist. We supply the technical evidence, you carry the responsibility for the use.
Who this is for
Defence, security, justice, health, European institutions, central government, operators of vital importance. Those for whom exposing a piece of data to a foreign jurisdiction is not a theoretical risk but an event to be prevented. For them, the hosting mode is not a cost parameter, it is a condition of operation.
Companies, local authorities, associations and foundations that refuse to let their internal exchanges, their deliberations and their intellectual property feed a service whose path and use they do not control. The issue is not only accidental leakage, it is the use the provider reserves for itself by contract.
System documentation, traceability of training data, risk assessment, governance. Open building blocks make these obligations demonstrable rather than declarative: you do not produce a statement, you show the code, the weights and the method.
How this is contracted
Five stages, the first of which commits little and clarifies a great deal.
- 01
Scoping
Uses, volumes, data sensitivity, regulatory constraints. Choice of deployment mode and of key custody.
- 02
Qualification
A qualification environment on a real use case, measuring performance and cost before any production commitment.
- 03
Go live
Deployment, data migration, integration with your directory and your tools, training for your teams.
- 04
Operations
Monitoring, updates, incident handling, evolution of the models served. You are informed before any model change.
- 05
Periodic review
Review of actual usage, sizing and costs, and reappraisal of the trade-offs in light of what has moved since.
Billing and commitment period
Usage-based or fixed-fee billing depending on the deployment mode chosen. Short, renewable commitment periods. We do not practise the multi-year non-terminable commitment, because a commitment you cannot leave is not a service, it is a dependency.
Price schedules are established during scoping, according to deployment mode and volume.
Let us talk about what you need to run
If you first have to get out of an existing contract, start with the dependency and reversibility audit.