Storage
Where do the sources and the prepared data live?
Sovereign AI platform
A sovereign AI platform is decided along the whole path.
Where is the content stored? Which extracts are sent out? Which model handles them? A sovereign AI platform is recognised by those answers.
Four levels to tell apart
Four levels to tell apart in a sovereign AI platform.
Preparation
Which services analyse, transcribe or index the content?
The model
Where is the answer generated, and under what retention terms?
The destination
Who receives the result, inside the platform or in a connected tool?
Choosing a configuration
Match the path to how sensitive the data is.
| Approche | Ce qu’elle permet | Ce qu’il faut vérifier |
|---|---|---|
| External model | Accéder aux modèles des fournisseurs disponibles. | Extraits transmis, conservation, réutilisation et localisation des traitements. |
| Locally hosted model | Rechercher une maîtrise plus directe de la génération. | Préparation des documents, indexation, journaux et tous les services annexes. |
| Path with protected data | Réduire l’exposition de certaines informations identifiantes. | Détection, données résiduelles, réidentification et limites de la protection. |
Organising the uses
A sovereign AI platform is built in from the start.
Separate the scopes
Build bases that match the audience, the departments and how sensitive the content is.
Set the access rights
Grant the rights the work requires, and review them as teams change.
Limit the data
Send only the information the task needs, and look into protecting personal data.
DécouvrirFrance & Quebec
Let us talk about a sovereign AI platform, starting from what you need.
Hosting options and processing requirements have to be pinned down for your project. Your context determines the level of control expected.
À examiner lors de la démonstration
- Lieu de stockage et services de traitement.
- Fournisseurs et sous-traitants mobilisés.
- Accès, conservation, suppression et restitution.
- Conditions contractuelles adaptées au projet.
Frequently asked questions
What you want to know about a sovereign AI platform.
Does Narratheque guarantee automatic compliance?
No platform replaces the analysis of your own processing or your organisation’s obligations. The features and the configuration have to be examined in your context.
Does multi-LLM protect data on its own?
No. It offers a choice. Protection depends on which models are allowed, which data is sent and the terms of each processing operation.
How do we look at a sensitive case?
Start by describing the categories of data and the result you expect, without sending sensitive documents. The path and the guarantees can then be framed.
Who decides the level of protection?
You do, corpus by corpus. Product documentation and a client file do not call for the same rules. The choice is made when the base is created, with a named owner, and is reviewed when the scope changes. A decision taken once for the whole site always ends up too loose somewhere and too strict elsewhere.
Let us start simply
Let us choose the framework for your sovereign AI platform.
Let us define your requirements before selecting the models and the connections.
