Copyright © 2022-2026 basebox GmbH, all rights reserved.
Licensed to be used in conjunction with basebox, only.

// admin

Models per app

Besides the organisation's default model, each app can set its own model. This is also how you control who uses a particular model: there is no separate set of model permissions; the way is an app with a fixed model and its app approval. This page also explains which model parameters you influence in the interface and which the Platform Operator sets.

Not every app needs the same model

Cost: A large model consumes more tokens per answer and, on basebox Server, ties up GPU capacity. It should only answer where it is needed.
Suitability: A model for analyses is unnecessarily slow for short answers.
Pilot operation: A new model should first be tested by a pilot team before it becomes the default model.

Models are set in three places

Model per app: "All apps" → Edit app → Advanced Fields → Model field (preset Use the default) and Thinking effort field.
Organisation default model: Administration → Organisation, Default model section, see Default model.
Inference (Server): the Platform Operator's Helm values, see Configure models.

As an administrator you can edit every app and view its model, including apps by other creators.

How basebox assigns models

Level	Setting	Effect
Organisation	Default model	Applies to all chats and to apps set to Use the default
App	App editor → Advanced Fields → Model	A fixed model for exactly this app
User	–	Users do not choose a model in the chat. In their own apps they can choose via the Model field; they cannot bypass a restriction to a group that way

Setting a model per app

Restricting a model to a group

Choose as the organisation's default model the model everyone should use.
Create an app (or use an existing one) and set the model only the group should get under Advanced Fields → Model. Adjust the Thinking effort if needed.
Share the app via app approval with the desired group only.
Leave all other apps on Use the default.

Result: the model answers only in this app, and only group members see it. The normal chat and all other apps continue to use the default model. It works the same the other way round: an app shared with the whole organisation with a fixed small model for a bulk task.

Tuning an app to its purpose

Open the app in the app editor.
Model: leave Use the default unless the app needs a specific model.
Thinking effort (for models that support it): Low for uniform tasks such as rephrasing or translating, higher for analyses and multi-step tasks. More effort means longer response time and more tokens.
App instructions: specify answer length, format and tone. This often has more effect than any parameter.
Save and test in the preview.

Which parameters are set by whom

Level	Who	What
Inference connection (Server)	Platform Operator, via Helm values	Endpoint, model identifier, context size, output limit, defaults for temperature, top-p, repetition penalty
Application	Administrator and app creators, in the interface	Default model, model per app, thinking effort per app, instructions

In basebox Cloud, basebox handles the inference level entirely.

Context size – how many tokens a chat can hold in total (questions, answers, documents). A property of the connected model, set on the server via AISRV_LLM_CONTEXT_SIZE; the Version Info page in the user menu shows the active value. For users: Context window.
Output limit – maximum length of an answer in tokens (AISRV_LLM_MAX_TOKENS, default 8000). An upper bound, not a target.
Temperature – variability of answers (0 = focused and reproducible, 1 = creative). Server default in AISRV_LLM_TEMPERATURE; cannot be set per app in the interface. Via the API, developers set temperature per request.
Repetition penalty / top-p – fine-tuning of the inference (AISRV_LLM_REPETITION_PENALTY, AISRV_LLM_TOP_P), set by the Platform Operator.

Details: AISRV → LLM configuration.

Note

Instructions have more effect than parameters. "Answer in at most five sentences" in the app instructions is more effective than a low output limit.
The organisation's system prompt applies in every app as well.
API keys can use all models listed for the organisation under /v1/models; there is no restriction per key. See API keys.
Consumption per app is shown in the Dashboard (tokens, users, amount).
Limits per organisation are under Organisation → Limits.
With an external inference endpoint, the model identifier configured in AISRV must exactly match what the endpoint reports under /v1/models.

Frequently asked questions

Can I assign a model directly to a group? Not as a separate setting. The way is an app with a fixed model, shared with the group.

Can users choose a different model in the chat? No. The model is determined by the organisation or the app.

What happens if a model that an app uses as fixed model is no longer offered? Switch the app to an available model or to Use the default.

How do I find out which apps deviate from the default? Open the app in the app editor and check Advanced Fields → Model. There is currently no overview.

Can I increase the context size as an administrator? No. It is a property of the connected model and is set on the server by the Platform Operator.

Can I set the temperature per app? No, not in the interface.

Next step: App approval · Dashboard

Need help? Contact support