Licensed to be used in conjunction with basebox, only.
// start
Start with basebox Server
Applies to
Product: Server · Audience: Platform Operator · Administrator · Procurement
basebox Server is a server for one organisation only. The hardware can be sourced through FAST LTA or procured by you (platform only); the server stands in your data center or in the basebox data center and is operated by you or – if commissioned – by basebox.
Before you install, you make four decisions. None of them changes the product; together they determine which pages of this documentation apply to you.
Decision 1: Hardware
| Option | What it means | Read more |
|---|---|---|
| Source through FAST LTA | Server class 1–3 for about 200 to 2,000 people, listed by basebox as reference configurations (class 1: 3 × RTX PRO 6000, class 2: 2 × H200, class 3: 2 × H200 + RTX PRO 6000; each with 128 GB or 512 GB RAM). Delivery three weeks after the order is received, installation in one day; hardware service contract with FAST LTA | Hardware options |
| Procure yourself (platform only) | Existing or self-procured hardware, including an existing cluster. basebox checks free of charge beforehand whether the performance is sufficient; licensing is by GPUs. If the hardware matches a reference configuration, that configuration's status applies; otherwise it is Custom | Reference configurations |
Keep in mind: the GPUs serve the service models and inference, not the basebox platform itself. How much GPU you need depends on the chosen language model, context and concurrency – see Understand the architecture.
Decision 2: Location
- Your data center – you control physical access, network and environment. The server can run fully air-gapped.
- basebox data center (hosting) – the server stands in the data center in Germany, by default at noris network in Munich; physical access and network connectivity are basebox's responsibility, the server still belongs to your organisation alone.
A server hosted at basebox is still Server, not Cloud. Details: Hosting options.
Decision 3: Operation
- Self-managed – after handover of the software you are responsible for the operating system, drivers, Kubernetes, basebox, service models, inference and network. basebox provides software, updates, documentation and support.
- Operated by basebox – if commissioned separately, basebox installs and/or operates the system via PAM/VPN or another agreed access path.
Buying the software is not an operations contract
A basebox Server licence is a software licence. You buy the server once; it belongs to you and is not part of the licence. The software comes as an annual licence (paid every year for as long as you use basebox) or as a perpetual licence (paid once, no time limit; updates and support from the second year are optional). Operation of the infrastructure by basebox is a separate engagement. What it includes is described under Operating models and Responsibilities.
Decision 4: Installation path
- From bare metal – you start with an empty server: operating system, NVIDIA drivers, Kubernetes, then basebox. The end-to-end path is the Bare-metal installation.
- Existing Kubernetes – you already have a cluster with GPU nodes, ingress and storage. Then you go straight to Install basebox.
Both paths converge at the Helm installation: Installation paths.
Checklist before you start
- Hardware decided and checked against a reference configuration
- Location and operating model settled contractually
- Network access to the basebox image registry (
gitea.basebox.health) or a plan for offline image transfer – see Air-gapped environments - Domain, DNS and TLS approach decided (cert-manager, your own certificate or local trust) – see Networking
- Language model chosen and sized against the GPU equipment – see Models & inference
- Identity integration planned (local users, LDAP/Active Directory, OIDC, SAML) – see Keycloak LDAP integration
- Person responsible for ongoing operations named – see Operations
After the installation
Once the installation is validated, things continue as in the Cloud: the first administrator signs in, sets up the organisation and invites users. The steps are under Administer – Getting started. The Platform Operator reads the first administrator's credentials from the Kubernetes secret basebox-admin-secret after the installation.
Frequently asked questions
Can inference run on a different server than basebox? Yes. The basebox platform can run on a CPU-only server and talk to a separate GPU system for inference. If the AI models come from a location other than basebox, this deployment option is called hybrid. See Deployment topologies.
Does the server need internet access? For installation it needs access to the registry and package sources, or mirrored artefacts. In operation, fully air-gapped use is possible; web search and online model downloads are then unavailable.
Who creates the first administrator? The installation creates it automatically (bootstrap job). The Platform Operator reads e-mail and password from the secret and hands them to the administrator.
Next step: Operate → Plan the server