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.
// reference
Glossary
Applies to
Product: Cloud · Server · Audience: All
The terms this documentation uses – and, where it matters, the terms it deliberately avoids.
- basebox Cloud
- Production environment on servers that basebox operates in the noris network data center (Munich East); the hardware is provided by FAST LTA. Own, separate environment with its own user management per organization, operated by basebox. No server owned by the customer.
- Transition phase
- Short for "basebox Cloud (transition phase until November 2026)". Cloud accounts created before the Cloud start run until the move in a shared environment at Hetzner in Nuremberg with AI models from Google Vertex in Frankfurt, logically separated from each other. See Transition phase.
- basebox Server
- A server for one customer only. Hardware from FAST LTA (class 1–3) or customer-procured; located in the customer's or basebox's data center; operated by the customer or, where commissioned, by basebox. Hosting at basebox does not make it Cloud.
- Platform only
- basebox on hardware of your choice – existing or self-procured, including an existing cluster. Licensed by GPUs. In this documentation a form of basebox Server.
- Hybrid
- basebox runs in your own network, the AI models (inference) come from another location. In this documentation a form of basebox Server.
- AI model in basebox / app with an external model
- basebox does not develop its own AI model; it runs open-source models itself – in the Cloud on basebox servers, with Server on the customer's hardware. Optionally an organization sets up an app with an external model that is visible only to the groups it is shared with.
- FAST LTA
- basebox's hardware partner: builds and delivers server classes 1–3 and provides the hardware service contract; also provides the Cloud servers. A hardware source – not a product, not a deployment model.
- basebox platform
- The management/application layer: UI, apps, authentication, permissions, audit, APIs, connectors, orchestration. Does not need large GPU resources.
- Service models
- Additional AI services used by basebox features: embeddings, RAG processing, OCR, STT, TTS, long-term memory. Dedicated GPU capacity is recommended.
- Inference
- The LLM/VLM/reasoning-model layer. Separate from basebox itself; sizing depends on model, quantization, context, concurrency and latency targets.
- Reference configuration
- A concrete, known hardware setup with documented topology, allocation and tested models. Not a requirement. Replaces the older terms hardware profile and hardware requirement.
- Validated / Supported / Experimental / Custom
- Status of a configuration or model: exactly tested by basebox · supported though not every combination recently benchmarked · expected to work, not production-validated · customer-specific, needs individual verification.
- Self-managed / basebox-managed
- Who operates a Server: the customer, or basebox under a separate commission (access via PAM/VPN or as agreed).
- Connector (MCP)
- An MCP server that gives the assistant a tool – web search, a wiki, a ticket system. basebox calls access through connectors a connection (German: Anbindung). Enabled per organization and per app; users sign in with their own account where required.
- Knowledge base / RAG
- The documents behind an app, searched before answering (retrieval-augmented generation). Answers cite sources.
- Terms this documentation avoids
- On-Prem / On-Premises as a product name (say Server, then state location and operation) · Hosted / Self-hosted as products · Hardware profile / hardware requirement (say reference configuration) · dedicated server / own server (say Server or a server for your organization only) · premium model, basebox model (say AI model in basebox or app with an external model) · integration for access through connectors (say connection).