For your tech team
The detail your CTO, CISO and DPO will ask for.
The rest of the site talks about outcomes. This page is the technical proof underneath: where it runs, how it is secured, how it is tested, and what we can and cannot claim today.
Deployment and regions
Your cloud
We build and run agents inside your own tenant or data centre. Your network rules, your identity provider, your logging.
Single-tenant
A dedicated environment operated by us, in Azure Switzerland North (Zürich), AWS eu-central-1 (Frankfurt) or GCP europe-west6 (Zürich).
Shared platform
Multi-tenant in Swiss and EU regions, for smaller teams and first projects. The same controls, a lower entry cost.
Security and data protection
Model gateway
Every model call leaves through one gateway: routed per task, logged, rate-limited and redacted. Personal data is removed before a prompt leaves your systems.
Retention
Zero data retention with the model providers that offer it. Where a provider does not, regulated workloads run on EU-hosted open models instead.
Agent permissions
Agents act through scoped tools with least-privilege credentials. Prompt-injection checks on inputs, policy checks on actions, an enterprise MCP gateway for tool access.
Encryption and testing
AES-256 at rest, encrypted in transit. Security and prompt-injection testing before every release that touches production.
How we test, and who decides
Golden sets with your experts
Evaluation sets written with your domain experts, plus adversarial suites and policy checks. Results ship with every version.
Human oversight by risk class
The EU AI Act risk class and your own risk appetite decide where a person approves, reviews or observes. We design that loop before the agent.
Traced end to end
Every prompt, response, tool call and cost is traced, so a decision can be followed back to its inputs and the version that made it.
Durable workflows
Long-running work survives restarts and retries, with every step recorded, so nothing is done twice and nothing is silently dropped.
Compliance status, stated plainly
What we hold, what we work to, and what is still in progress.
- GDPR and Swiss nDSGBy designRecords of processing and a DPIA per use case, maintained with your DPO.
- EU AI ActMapped per use caseRisk classification, transparency duties under Article 50 and technical documentation templates for every deployment.
- ISO 27001Aligned, certification in progressOur controls follow ISO 27001. We will say so here once the certificate is issued.
- SOC 2Not certifiedWe share our control set and answer your security questionnaire instead.