ISO/IEC 27001
Our company ISMS is certified. Extending the scope to this service is planned.
Client data, unpublished findings and exploit research belong in a private environment. Every instance runs on a dedicated GPU in the EU, stores no prompts or outputs, and is never reached through our control plane.
Nine design principles, in three areas.
Where your instance runs and who operates it.
Each instance hosts one model on its own GPU. GPUs are not shared between customers.
Instances run only with EU-based GPU providers that are certified to ISO/IEC 27001. You do not operate or configure the provider.
The company behind the service runs an ISO/IEC 27001 certified information security management system. Extending the certification scope to this service is planned.
What happens to prompts, outputs and instance content.
Prompts and outputs are not used to train models.
Prompts and outputs are never stored. Request logging is off, and nothing is written to logs or disks.
Stopping an instance deletes it. Instance content is not kept afterwards.
Who can reach your instance and what is agreed.
Our control plane can start, stop and delete your instance. You call the instance directly over TLS, and inference traffic never passes through the control plane.
Each instance has its own API key, checked on the instance. Only a hash of the key is stored by us.
An AVV / DPA is part of onboarding, together with a documented list of subprocessors.
The control plane manages the lifecycle of your instance. It is never part of the data path.
From the request to the deletion of the instance.
What is held today and what you receive when you onboard.
Our company ISMS is certified. Extending the scope to this service is planned.
All EU GPU providers are certified to ISO/IEC 27001. Further certifications such as SOC 2 are listed per provider in the data processing agreement.
An AVV / DPA and a documented list of subprocessors are part of onboarding.
The models are built for authorised security work. Some Red Team models need verification before you can deploy them, and every use remains subject to the terms and applicable law.
On your own instance: one model on a dedicated GPU, with an EU-based GPU provider. Requests go straight to the endpoint of your instance.
No. Inference traffic does not pass through our control plane, prompts and outputs are never stored, and we have no technical access to running instances. The data processing agreement is the binding reference.
No. Prompts and outputs are not used to train models.
Stopping an instance deletes it. Instance content is not kept afterwards.
It supports GDPR-oriented processing, but compliance also depends on contracts and on how you use the service. That is why a data processing agreement is part of onboarding.
The data processing agreement and the terms of service are binding. This page describes how the service works.
Tell us about your team and the models you want to run. We will contact you as soon as your tier is available.