Skip to content

// installation

Customer-managed operations

Applies to

Product: Server · Operation: by the customer · Audience: Platform Operator · IT management

What the customer is responsible for after handover of the software: operating system, drivers, Kubernetes, basebox, service models, inference, network and backups. basebox provides software, updates, documentation and support – you operate the server. This page is your task list.

What you are responsible for

Area Your task Runbook
Physical hardware, data center Power, cooling, network connectivity, access; with hosting at basebox, basebox takes this row Hosting options
Operating system Security updates, hardening, access (SSH keys, no root login, fail2ban, ufw) Server Preparation Guide → Security
NVIDIA driver, CUDA, GPU Operator Version state, compatibility with vLLM, MIG layout after reboots NVIDIA / GPU
Kubernetes Cluster health, certificates, ingress, storage, CloudNativePG, upgrades Kubernetes
basebox Helm releases, values, updates, migrations Updates
Service models ragsrv/ragsrv-support healthy, model caches, GPU/MIG allocation Deploy service models
Inference Endpoint reachable, model loaded, latency, capacity Connect inference · Models & inference
Network DNS, TLS renewal, firewall, egress for integrations Networking · VPN / networking
Backups Databases, media, secrets; restore rehearsed Backup & restore
Monitoring and logging Alerts, log collection, retention per your policies Monitoring · Logging
Connector workloads Deploy MCP services via Helm, egress to target systems MCP connectors with Helm
Identity Keycloak: LDAP federation, SSO, admin access LDAP / OIDC

Not your task as Platform Operator: application administration. Users, apps, model selection, connector enablement and policies lie with the administrators – even if the same person does both, they are separate roles.

What basebox provides

  • Software: Helm charts and container images from the basebox registry; versions in the Version matrix.
  • Updates: security and feature updates, announced by e-mail and release notes; no automatic updates – installation is up to you.
  • Documentation: these pages, the compliance package, release notes and changelog.
  • Support: support@basebox.ai per the contractually agreed response times; for cases this guide does not cover, for air-gapped installations and custom configurations.
  • Notification: in case of known systemic risks or security-relevant incidents, basebox informs affected customers without delay.

What basebox does not do

Without a separate engagement, basebox has no access to your server – no remote maintenance, no monitoring, no updates by basebox. If a support case needs access, that happens only after your approval via an agreed path (PAM/VPN, time-limited) – see Remote maintenance.

What your team should bring

  • Linux administration (Ubuntu), package management, systemd, firewall.
  • Kubernetes day to day: kubectl, helm, pods, PVCs, ingress, secrets, reading events.
  • NVIDIA stack: driver updates, nvidia-smi, GPU Operator, MIG if needed.
  • PostgreSQL basics for backup and restore (CloudNativePG).
  • Willingness to read release notes before updating.

If that is not available in-house, Operation by basebox is the path – or a service provider of your choice.

Handover from basebox to you

If basebox installs and you operate afterwards, the handover should include:

  • Manifest: chart/app version, image digests, driver, CUDA, Kubernetes, GPU Operator, GPU UUIDs with role, MIG layout
  • Values files (without secrets) and the list of Kubernetes secrets
  • Credentials: Keycloak admin, first basebox administrator (already handed to the administrators)
  • Acceptance record per Validate installation
  • A first backup and evidence of a rehearsed restore
  • basebox's access is terminated (PAM/VPN disabled) unless an operations engagement follows

If you get stuck

  1. Troubleshooting and FAQ.
  2. Have logs of aisrv and the affected component, events and the manifest ready.
  3. support@basebox.ai with symptom, time, recent changes – without credentials. For security incidents additionally datenschutz@basebox.ai.

Next step: Monitoring