Copyright © 2022-2026 basebox GmbH, all rights reserved.
Licensed to be used in conjunction with basebox, only.
Licensed to be used in conjunction with basebox, only.
// installation
basebox Server installation
Applies to
Product: Server · Audience: Platform Operator
This is a legacy entry point
This page keeps its URL for existing links. The Server documentation is now organised as an end-to-end path under Developer → Installation & Operations → Plan the server. Start there if you are new.
basebox Server is installed on Kubernetes with the basebox Helm chart. docker-compose is possible but not officially supported.
Choose your path
| You have … | Start with |
|---|---|
| an empty server (bare metal) | Bare-metal installation – requirements, OS, NVIDIA, Kubernetes, storage, network, install, validate |
| an existing Kubernetes cluster | Using Helm Charts – public domains, cert-manager or customer-managed TLS |
| a local evaluation machine | Local Quick Start – basebox.local with local/self-managed TLS |
| basebox and inference on separate hosts | Deployment topologies – CPU-only application server + external OpenAI-compatible inference |
Before you install
- Understand the architecture – platform, service models and inference are three layers; only the last two need large GPUs.
- Reference configurations – tested hardware layouts (2 × H200, 4 × H100, RTX PRO 6000 systems).
- Responsibilities – what you own in self-managed operation, what basebox can take over under contract.
Reference material on this legacy path
- Server Preparation Guide – Ubuntu, NVIDIA drivers, CUDA, container runtime, Kubernetes prerequisites
- Helm chart overview and per-service configuration
- Connecting Keycloak to LDAP / Active Directory
- LLM recommendations (sizing)
- FAQ
Legacy terminology
Older pages and configuration keys on this path use "on-premise" (e.g. AISRV_ON_PREMISE, STORESRV_ON_PREMISE, and this URL). Read it as basebox Server: a dedicated server for one customer, in the customer's or basebox's data center, operated by the customer or by basebox. See Deployment models.