Sovereignty, level by level.

“Sovereign” is an easy word. We prefer to be precise.

A data and AI solution has several layers. Each one can be local, or not.

Here is our position, layer by layer.

Six levels

For each layer, we say whether it can stay local, and under what conditions.

Six levels
Level	What it is	Status	Our position
Storage	Where the data is physically stored.	Status : Can be local	Data can stay in Morocco, with the organization or with a local host. We use open formats so that it stays portable.
Query engine	The software that reads the data and runs the computations.	Status : Can be local	We favor open source engines. They run where the data is.
Catalog and permissions	What describes the data, and who is allowed to access it.	Status : Can be local	The catalog and access rights can be hosted locally with open tools. This is the layer that makes everything else governable.
Embeddings	The models that turn text into vectors, for information retrieval.	Status : Depends on the project	Open models can run locally. Their quality in Darija and Amazigh still has to be measured. We test and we publish.
Language model	The model that reads, summarizes, reasons and answers.	Status : Often external	The most capable models are often offered by foreign providers, through an API. Open models can run locally, with a cost in hardware and sometimes in quality. We choose case by case, and we say so.
Operations	Monitoring, updates, security and day-to-day support.	Status : Depends on the project	It can be run by a team in Morocco. It also depends on software and hardware designed elsewhere.

Position updated in October 2026. It will evolve with our tests.

We do not say “fully sovereign”.

We do not promise that no data ever leaves.

Hardware, much open source software and the most advanced models often come from elsewhere.

We say what is local, what is not, and why.

Share your tests and measurements with the circle.