See itbeforeyou trust us.
One place that shows where your data lives, what we commit to, who our suppliers are, and which documents you can request. We tailor data residency to your requirements: where it matters, data and inference stay inside the EU throughout. Compliance is the starting point of the design, not a later add-on.
Four areas, with concrete commitments
Not principles, but verifiable commitments. We can show each of them on a call or in a document.
The data is yours, and it stays in the EU on request
- Where compliance requires it, data and inference stay in the EU.
- GDPR-based data handling, with a data processing agreement.
- We do not train on your data, and beyond our sub-processors we do not pass it to third parties.
- Deletion on request, with a predictable retention period.
Transparent AI, not a black box
- Every answer names its source.
- EU AI Act readiness, with the obligations built into the design.
- A full audit log of every step.
- Typed tool use, human override, a strict stop condition.
Audited before it goes live
- The system passes the security audit before going live.
- Zero-trust access, secrets scanned out of the pipeline.
- CodeQL and a security review already in the Pull Request.
- Real-time threat detection.
The live system is not left on its own
- Monitored operation, with alerts.
- Incident handling and RCA after every disruption.
- Predictable, audited releases.
- Backups and a documented recovery plan.
Who can access the data
The infrastructure and model providers the systems may run on. What is actually used comes down to your setup: from a fully on-prem or internal model (with no external sub-processor), through EU-region endpoints, to global APIs. The complete, current list is in the DPA annex.
On-prem / internal
The model runs on your infrastructure. The data never leaves the environment, and there is no external sub-processor.
No external partyEU-resident
EU-region providers and endpoints only. Data and inference stay inside the EU throughout.
Inside the EUGlobal, task-matched
The best model for the task, with no residency constraint, across a multi-provider gateway.
Mixed region| Provider | Role | Region |
|---|---|---|
| OVHcloud | Hosting and infrastructure | EU (France, Germany) |
| Netcup | Hosting and infrastructure | EU (Germany) |
| Cloudflare | CDN, WAF and DDoS protection | Global edge, with EU data handling |
| Model inference (EU region) | Mistral, and the hyperscalers' EU-region endpoints (Azure, AWS, Google, OpenAI EU) | Inside the EU |
| Model inference (global) | Task-matched model, on request: OpenAI, Anthropic, OpenRouter, HuggingFace, Alibaba, z.ai, Kimi, MiniMax | Mixed region, only without an EU residency constraint |
The AI layer is a multi-provider gateway, not tied to any single model supplier. For an EU residency commitment we use EU-region providers and endpoints only; for a fully on-prem setup the model runs in your environment, with no external sub-processor. The current, named list of model sub-processors is in the DPA annex, and we send it on request.
What you can request
The details are not public, but we send them after a call or a signed NDA.
What we get asked most
The short answers here; the details on a call or in the documents.
Where does inference run, and where does the data stay?
Do you train on our data?
Who works on the project, and who do we get?
How long, and what does it cost?
What if we want to move off you?
What models do you use?
Ask about security directly.
For a compliance or security question too, you speak with an engineer, not a salesperson. We reply within one working day.