Confidentiality
Your customers', patients' or employees' data is processed where it was created. Nothing passes through a third-party service.
Local AI · on-premise
For organizations whose data is sensitive or voluminous, we deploy language, vision and speech models directly on your infrastructure. No per-request subscription, no data leaving.
Your customers', patients' or employees' data is processed where it was created. Nothing passes through a third-party service.
An amortized server replaces thousands of per-token requests. Marginal cost tends to zero.
Works offline, no quota, no forced version change.
In Morocco, sending personal data abroad is a transfer subject to authorization (arts. 43-44, Law 09-08). Locally, the question doesn't arise.
GPU and memory choice according to the model, the precision and the real load. We publish a free tool for that.
Open the GPU tool →Open models evaluated on your real cases — French, Arabic, Darija — and quantized to fit the available hardware.
Installation, inference service, monitoring, updates. On a workstation, dedicated server or cluster.
Indexing of your documents, sourced answers, access rights respected. The corpus stays in your network.
Internal API, ERP/CRM connectors, business interfaces.
Processing description, security measures, impact assessment and information notice, ready for your formalities.
| Criterion | Cloud API | Local model |
|---|---|---|
| Data location | At the provider, often abroad | On your premises |
| Cost | Per token, grows with usage | Amortized hardware, near-zero marginal cost |
| Availability | Depends on a third party | Works offline |
| Frontier models | Best proprietary models | Open models highly competitive on targeted tasks |
| Transfer formalities | CNDP request or authorization | None |
We are not dogmatic: when an API is the right choice, we put it behind a filter that strips identifying information before the call.
Let's discuss how AI can revolutionize your business processes and drive growth.