Licensed to be used in conjunction with basebox, only.
// admin
Enabling models
Overview
Which AI models are available to your organisation is set here. This page complements Model selection with the question of where the list of models comes from – and the answer differs between basebox Cloud and basebox Server.
What it is for
The model list determines what your users can select in apps, what the API returns under /v1/models and which model is a candidate for the default. A short, vetted list is clearer and more cost-predictable than a long one.
Where the models come from
| basebox Cloud | basebox Server | |
|---|---|---|
| Who connects models | basebox operates inference and provides the models | The Platform Operator connects the application server (AISRV) to one or more inference endpoints |
| What you choose from | The models provided by basebox | The models reported by the connected endpoint (with vLLM, all models loaded there) |
| An additional model | Request to basebox | The Platform Operator loads it into inference or connects another endpoint |
On a server, the model list is therefore a matter of infrastructure: an additional model needs GPU memory and is planned under Models & inference. A model from the internet (external provider) is a data protection and contractual question – see Model providers.
Where to find the setting
In your organisation's administration area, together with model selection.
Which model is currently active and what context size it has is shown on the "About" page in the user menu.
Step by step
Provide a model for the organisation (Server):
- The Platform Operator checks the model against the GPU equipment (LLM recommendations) and loads it into inference or configures the endpoint in AISRV (
AISRV_LLM_URL,AISRV_LLM_MODEL,AISRV_LLM_CONTEXT_SIZE). See Configure models. - Check that the model appears under
/v1/models(a developer can test this in seconds with an API key). - Choose it as the default model if needed, or assign it to individual apps.
- Test in a private app before releasing it widely – answer quality, speed, reasoning support, tool calling for connectors.
Withdraw a model:
- Switch apps that use it as fixed model to another model or to "Use default".
- Choose another default model if it was the default.
- The Platform Operator removes it from inference.
Notes
Note
- Not every model can do everything. Reasoning (thinking modes) and tool calling (connectors) require models that support them. Models without reasoning show no reasoning selector in the chat.
- Context size is a model property. It determines how much text a chat can hold – see Context window.
- Known model quirks (such as Qwen 3 with
/no_think, DeepSeek-R1 with thinking tokens, GPT-OSS with tensor parallelism) are under LLM recommendations → Known issues. - No silent fallback: if a model's inference endpoint is offline, basebox shows the error; it does not switch to another model unnoticed.
- The compliance package under the EU AI Act lists the models shipped by basebox in the Model Register.
Frequently asked questions
Can I offer several models at the same time? Yes. The default model applies wherever nothing else is set; individual apps can set a different model. On basebox Server the models have to be connected for this – see Multiple inference instances.
Why do I see different models in the Cloud than a colleague on a server? Because the list depends on the respective inference. In the Cloud basebox provides it, on the server the Platform Operator does.
How do I tell whether a model supports reasoning? By the reasoning selector in the chat: if it is missing, the active model does not support reasoning.
Who decides on external model providers (OpenAI, Anthropic)? Your organisation – contractually and under data protection law. Closed-source models are not part of the basebox delivery; see Disclaimer.
Need help? Contact support