Security &
Compliance First

noucore follows a strict zero-data-leakage policy. Because the architecture is operated fully encapsulated in your controlled environment (either in a dedicated, ISO 27001-certified EU data center or on-premise), data never leaves your network. Since we use open-weight models exclusively, there are no third-party APIs through which data could be misused for model training.

Yes. noucore relies on role-based access control (RBAC) and connects seamlessly to your identity provider (e.g. Microsoft Entra ID, Keycloak) via OAuth2, OpenID Connect (OIDC) or SAML 2.0. noucore only reads and processes documents from your source systems (e.g. ERP, CRM, file servers) for which the requesting user has explicit read permission.

Yes. With an on-premise deployment in your own data center / server room, the entire AI stack – including the user interfaces, inference engines and vector databases – can be operated in air-gapped mode. No active internet connection is required for regular operation or for inference of the open-weight models. This makes noucore ideally suited for critical infrastructures (KRITIS).

Local or dedicated EU hosting eliminates the risk of violations relating to transatlantic data transfers (such as the US Cloud Act). All data processing steps are transparent and fully auditable for your compliance department. In addition, by using open-weight models and local vector stores, noucore is ideally prepared for the documentation and transparency requirements of the EU AI Act.

noucore features standardized exporters for Prometheus that integrate natively into your Grafana dashboards. Both hardware metrics of the compute layer (GPU utilization, tensor core load, VRAM allocation and temperatures) and application-side KPIs (time-to-first-token, inference latencies, API throughput and HTTP error rates) are monitored.

Because the entire noucore architecture is containerized, lifecycle management is standardized via Docker infrastructures. Security patches for the operating system, updates to the inference software or the replacement of AI model weights are packaged as new container images and can be rolled out via your established CI/CD pipelines. With an appropriately redundant node design, this takes place during live operation via rolling updates without any system downtime.