No vendor access path
Hyground GmbH runs no SaaS control plane and has no operator access into customer deployments. No phone-home, no central key, no shared infrastructure. The vendor has nothing of yours to lose.
Security & Compliance
Governed Autonomy. Control Architecture First
Hyground is a self-hosted autonomous operations layer that runs entirely inside your perimeter. Scoped service identities with least-privilege access. Read-only by default. No autonomous mutation. Every output auditable. Zero data egress, zero telemetry, no external model training. Security is enforced at the architecture level.
Platform Hardening
Hyground is delivered as a Kubernetes-native Helm chart and runs entirely in your cluster. Defaults are tight: hardened images, scoped permissions, and a read-only cluster RBAC. The vendor has no operational path into your cluster.
Authenticated identity on every action
Sessions, queries, and actions tie to a named user via OIDC (Entra ID, Zitadel, any OIDC provider). Automation runs under scoped technical users with the same audit trail.
Secrets redacted before the model sees them
Tool responses pass through a secret filter before reaching the LLM. If an integration accidentally returns a credential, the model never sees it. Redaction happens in your cluster, before the response reaches the model.
Hardened container images
Chainguard distroless base images. Containers run as non-root with a read-only root filesystem, Linux capabilities dropped, no privilege escalation, seccomp applied. SBOM published with every release.
Kubernetes-idiomatic by default
The Helm chart ships NetworkPolicies, per-service ServiceAccounts, and a Kubernetes adapter ClusterRole limited to get, list, watch. Read-only enforcement reaches the cluster-RBAC layer.
Air-gap deployable
Air-gapped on-premises deployment supported. The Helm chart is designed for offline install. The only outbound traffic is to the LLM provider you choose.
Control Architecture
Every security property is enforced at the infrastructure level. Scoped identities, least-privilege access, and adapter-level read-only enforcement are structural, not configurable.
Scoped Service Identities
Hyground agents operate under scoped service identities with least-privilege access to your infrastructure. Each identity is bound to specific namespaces, data sources, and action types. No shared credentials, no ambient authority.
Read-Only by Default
All operations are read-only. Hyground investigates, analyses, and reports. It does not restart, scale, or reconfigure anything. Adapters enforce this at startup, below the LLM.
Zero Data Egress
No telemetry, no usage tracking, no phone-home behaviour. All processing happens within your cluster. The architecture physically prevents data from leaving your perimeter.
Read-only enforced at the adapter
Database adapters verify read-only roles at startup. CLI adapters block write commands and flags. Each integration is bounded by the adapter itself, below the language model. A misbehaving model cannot escalate beyond what the adapter executes.
Your LLM, Your Choice
Connect Azure OpenAI, Anthropic, Google Vertex, or run fully self-hosted with Ollama. You control the model provider. Your data is never used to train external models. The customer retains model choice at all times.
Auditable Outputs
Every investigation, every query, every finding produces a structured, exportable record. Compliance teams can audit exactly what the system did, what data it accessed, and what conclusions it reached, without engineering involvement.
See Hyground in action
Check out our sandbox or schedule a demo with our team and experience sovereign AI for DevOps firsthand.