AI Models
How to choose the AI model for each chat in Kasimir and how admins govern models, the default model and EU data residency.
Kasimir is a multi-LLM platform: you don't chat with a single fixed model — you choose, per conversation, which AI model answers. The default model is hosted by Kasimir itself, on its own servers in the EU, making it the most GDPR-sovereign option. Stronger cloud models can be added, while it always stays transparent where your data is processed. This page explains how to switch models in chat, what the badges and hints mean, and how administrators govern the available models for the whole company.
Choosing the model in chat
Every conversation uses exactly one model. The model switcher sits in the composer at the bottom of the chat: a button with a colored dot and the model name (e.g. Gemma 4 (31B, EU)). Clicking it opens the list of models enabled for your company.
Pick a model from the list — from the next message on, that model answers.
Your most recently used model is remembered in the browser and pre-selected the next time you start a new chat.
If your administration has set a default model, it is pre-selected in new chats as long as you haven't picked a model yourself. Your own choice takes precedence, and you can switch at any time.
Badges: own servers, EU cloud, non-EU
Each model in the list carries a badge that shows at a glance where processing happens:
Badge | Meaning |
|---|---|
| Processed on Kasimir's own servers in the EU — the most sovereign option (e.g. the default model Gemma 4). |
| Processed in an EU region by a cloud provider acting as a data processor. |
| Processed outside the EU — your chat data leaves the EU. These models are hidden by default and must be enabled by your administration. No offered model is currently classified this way. |
The detail panel (hover)
On desktop, hovering over a model in the list reveals a small info panel showing:
the provider (e.g.
Kasimir,Microsoft Azure,Anthropic,Mistral AI),a short description of the model,
Context— the size of the context window (e.g.128ktokens),Image understanding— whether the model can "see" images or PDFs (yes/no),Data processing— where your inputs are processed.
For Data processing, three tiers are distinguished:
Processed on Kasimir's own servers in the EU — the most sovereign option (default model).
Processed in the EU · Azure AI Foundry (Microsoft) — EU region, but via a third-party processor.
Processed outside the EU – data leaves the EU — only for Non-EU models.
Data-processing hint below the input
When you pick a model that does not run on Kasimir's own EU infrastructure, a hint appears directly under the input field — so you know where your data goes before you send:
EU cloud:
Processed in the EU via Azure Foundry (Microsoft)(neutral hint).Non-EU:
Note: processed outside the EU – data leaves the EU(amber warning).
For the self-hosted default model no hint is shown — here data never leaves Kasimir.
The context window (e.g. 128k) determines how much text — including attachments, documents and conversation history — fits into a conversation. The context meter next to the input shows current usage against the selected model's window.
Managing models for the company (Administration)
Owners and administrators control which models are available to members under Administration. Two areas matter here.
AI Models — turn models on and off
The Administration → AI Models page lists every chat model available to the company. Each row shows the model name, the provider, whether it runs Self-hosted or Cloud, and — if paid — the cost per 1k tokens (input/output).
The toggle on the right enables or disables a model for the entire company. Disabled models no longer appear in anyone's chat model switcher. Changes only take effect on Save; Discard resets them.
Below that you set the default image model – the model for image generation in chat and in the image generator.
Plan and quota: Models that don't run on Kasimir's own servers (badge EU cloud or Non-EU) depend on your plan. The Basic plan doesn't include them; the self-hosted models remain unlimited there. On the Advanced plan, each person has a quota of 750,000 tokens per calendar month for these models. Once it is used up, Kasimir rejects further requests to them until the end of the month.
Chat settings — default model and the EU gate
The Administration → Chat settings page sets the default chat behaviour for all members. Two options matter for models:
Default model— the model pre-selected in new chats. Choosing(no default — users pick themselves)leaves the choice to members. If someone has already picked a model themselves, their choice takes precedence; anyone can switch at any time.Allow non-EU models— the data-residency gate (see below).
The page also controls other chat defaults such as web search, file uploads and a company-wide system-prompt addendum.
The EU / GDPR gate for non-EU models
Kasimir's core advantage is EU-sovereign hosting. Models that would process data outside the EU are off by default — so privacy-sensitive companies never even see them. At present, all offered models run on Kasimir's own servers in Germany or in the EU cloud.
Only when the administration enables Allow non-EU models under Chat settings do these models (with an amber Non-EU badge) appear in the model switcher. While the gate is off, they are hidden everywhere — in the chat switcher as well as via the API.
Allow non-EU models enables models hosted outside the EU. Chat data then leaves the EU — this is not GDPR-sovereign. The option is off by default; enable it deliberately and only if your data-protection rules permit it.
The gate applies company-wide and on every path: even when a non-EU model is addressed via the programming interface (API), Kasimir rejects the request as long as Allow non-EU models is not enabled.