Navin Enterprise AI is a self-hosted, multi-tenant platform. The same platform can be deployed in five topologies, so that each organisation decides how far it isolates its data and its models.
Five topologies
- Single brick: the Control Plane, PostgreSQL and one component, such as the LiteLLM gateway, with API key authentication.
- Shared SaaS: one cluster, several tenants isolated by RLS, OpenFGA and Cilium namespaces.
- Dedicated tenant: a namespace or a cluster per large account, with its own OpenBao mount and GPU pool.
- Sovereign on-site: an air-gapped cluster, local models only (vLLM), a Harbor mirror and Argo CD from an internal Git.
- Hybrid: local vLLM for restricted data, remote providers allowed by policy for the other classifications.
Air-gapped by design
In the sovereign topology, inference runs on vLLM inside the cluster and LiteLLM remains the single gateway to the models. Images come from a Harbor mirror and Argo CD deploys from an internal Git repository. Every topology is delivered by Argo CD from Helm charts, with images signed by Cosign, Kyverno admission control and Velero backups.
Installing with OS packages
Outside Kubernetes, the three Navin products ship as DEB, RPM and pacman packages with systemd services, for Debian and Ubuntu, the RHEL family (Red Hat Enterprise Linux, Rocky Linux, AlmaLinux, Oracle Linux, CentOS Stream) and Arch Linux. A single installation script checks every prerequisite before the first change, in standalone and cluster mode alike. Offline archives, one per OS family, contain the packages, which are verified again at installation time.
The Navin Manager agent opens all of its connections itself, so managed hosts need no inbound port. Archives, images and Python packages of the offline mirror are verified by their sha256, and the vendor package repositories are signed offline by the release manager, never in CI.
Nothing changes while a check is failing.