Skip to content

// installation

basebox platform

Applies to

Product: Server · Audience: Platform Operator

The platform layer is everything a user or administrator touches directly, plus the services that hold basebox together. It is the part of basebox that is basebox – and it is deliberately light on GPU.

What belongs to the platform

Component Role
frontend The web UI
AISRV Application server: chat orchestration, apps, connectors (MCP gateway), OpenAI-compatible and REST API
storesrv Document and object storage service
Keycloak (IDP) Authentication, users, groups, LDAP/OIDC federation
Databases PostgreSQL (CloudNativePG) for platform state
Ingress / TLS Entry point for users and API clients

Component-level configuration is documented per service under Helm → Services; a consolidated component map is on basebox components.

Resource profile

  • CPU, RAM, storage, network – sized by number of users, documents and API traffic.
  • GPU: none required. The platform can run on a CPU-only application server. When it shares a node with GPUs, it does not consume them.

This is why a reference configuration's GPUs are always accounted to service models and inference, never to the platform.

Where it can run

  • On the same node as service models and inference (single-node reference configurations)
  • On a separate CPU-only application server, with inference on a dedicated GPU host – see Deployment topologies
  • In a Kubernetes cluster with mixed CPU and GPU nodes, where the platform is scheduled to CPU nodes

What the platform is not

  • It is not the model. Changing or upgrading the LLM does not change the platform.
  • It is not the OCR/STT engine. Those are service models.
  • It is not the hardware requirement. Statements like "basebox needs an H200" describe inference, not this layer.

Next step: Service models