Models and deployment
You choose where your AI runs.
The neuland.ai HUB runs in the EU cloud, in Schwarz Digits Cloud (STACKIT), in your data center or hybrid. You switch models without rebuilding and run your own model alongside them.
Muster AG · Location per department
Not every task may be processed anywhere. The protection level determines the location.
Confidential documents need more protection than ad copy. In the neuland.ai HUB, you choose the deployment model and approved models per department, from the EU cloud to your own data center.
Deployment models
Four deployment models, freely combinable: for the whole enterprise or per department, matching how your information security classifies documents.
Model choice
In Auto mode, each message goes to a suitable model based on its complexity: a quick follow-up question to a fast one, the summary of a 40-page customer contract to a strong one. Anyone who wants to can choose for themselves, even mid-chat. Existing contracts with model providers plug in via Bring Your Own Key (BYOK). All models compared
| Time | Who | Action | Model | Status |
|---|---|---|---|---|
| 09:41:07 | Legal | Customer contract summarized | Strong · in‑house | Auto |
| 09:43:52 | Legal | Liability cap question answered | Fast · in‑house | Auto |
| 10:04:12 | Procurement | Quotes for cast housings compared | Standard · Schwarz Digits Cloud | Auto |
| 10:18:45 | HR | Job ad drafted | Standard · EU cloud | Auto |
| 10:31:02 | Legal | Liability clause drafted to company standard | Custom model · in‑house | manual |
Legal · Customer contract summarized
Legal · Liability cap question answered
Procurement · Quotes for cast housings compared
HR · Job ad drafted
Legal · Liability clause drafted to company standard
- Legal and Engineering can only choose models in the corporate data center.
Model benchmark
Which model for what?
Measured independently.
How strong, how fast and how expensive common models are, measured by Artificial Analysis, plus which ones can run in your own data center. Go to the model benchmark
For example, these models, proprietary and open
OpenAI
Claude
Gemini
NVIDIA
Llama
Mistral
DeepSeek
Qwen
Kimi
GLM
Cohere
Aleph Alpha
Nemotron
Nemotron: open weights, optimised for NVIDIA hardware.
And exchangeable in the HUB at any time.
You can run, fine-tune and swap out NVIDIA's Nemotron model family in the HUB just like any other model. Four points, soberly assessed.
Training data
Largely synthetic
Nemotron was trained largely on synthetic data. The practical benefit lies in transparency: much of the data used has been disclosed, although not all of it.
Hardware
Optimised for NVIDIA
The models are tuned to NVIDIA hardware. On top of that comes a mature NVIDIA stack for fine-tuning that keeps the path from trial to production short.
Sizes
From small to large
The family ranges from small to large models. That lets you pick the model that is just large enough for each use case, instead of always the largest one.
Fine-tuning
On your own or with us
You can fine-tune Nemotron models yourself and then connect them in the HUB. On request we handle the fine-tuning for you, using your domain knowledge from your own documents.
Model switch
Agents, knowledge and orchestrations are not tied to one model. An example from Procurement: Monday to Friday, without anyone rebuilding anything.
Control
The rules travel with you, wherever the computing happens: switching the model does not switch the protection.
Location
You choose per department
EU cloud, Schwarz Digits Cloud, corporate data center or hybrid. You decide, not the model provider.
Training
Excluded
Your data, from customer contracts to engineering reports, is never used to train language models.
Permissions
Per role
Legal sees no HR files, HR sees no contracts, even in chat.
Log
Complete
Access and events are logged by the same rules at every location, even after a model switch.
FAQ
Which department runs where?
We define it with you.
You bring your departments; we assign each one a deployment model and its approved models, as the basis for your onboarding plan.
