Hosting in France, inference on a private network, self-hosted models, row-level isolation and a native audit log. An architecture built first to reduce your exposure to extraterritorial law. Without claiming to « neutralise the Cloud Act ».
Hosting and inference stay in France.
Your mandates are confidential by nature. The architecture is designed so that no criterion, no target and no intent ever passes through a service outside your perimeter.
- Hosting in France (OVHcloud)
- Inference on a private network (vRack, not exposed)
- Open-weight models, self-hosted in France
Your data feeds no third-party model.
The AI runs on open-weight models we serve ourselves: no closed API in the inference loop, no dependence on a provider that could change its terms.
- No data sent to OpenAI or Anthropic
- No re-training on your data
Every organisation is isolated, every access is logged.
Isolation is enforced in the database, not only in the application. The traceability a due diligence demands is native: you know who saw what, and when.
- Multi-tenant row-level isolation (RLS)
- Native audit log
- No cleartext access in normal operation (logged break-glass procedure)
What we display, and what we do not.
No SecNumCloud or ISO 27001 certification yet: the architecture is built along those lines (hosting in France, isolation, audit), but we display no label until it is earned.
For Enterprise deployments, a single-tenant instance or dedicated hosting can be discussed against your security requirements.