What does on-premise AI mean for you?
On-premise AI means the language model runs on infrastructure you control, so your data never leaves the building. There are no calls to a global cloud, no copies sent to a foreign provider, and no exposure to another country's jurisdiction. Od1n is a sovereign 3B Norwegian language model with its own weights that can be hosted inside the EEA or fully on-premise on your own hardware, with no data sharing to third parties.
For a buyer, that is the whole point: the answers come to your documents, your documents do not go to someone else's server.
How does it actually work? The four layers
Od1n is best understood as four layers stacked together. Each one is simple on its own.
- The model. A compact 3B Norwegian language model with its own weights. It understands and writes Norwegian well, and it lives on your infrastructure rather than in a global cloud.
- The knowledge base. Your own documents, procedures, contracts, case files and internal knowledge. This is what makes the system useful to your organisation, not a generic assistant.
- Answers from your sources (RAG). Retrieval-augmented generation lets the model look up the relevant passages in your knowledge base and answer from them, with the sources attached, instead of guessing from memory.
- Integration. The system connects to the tools and workflows you already use, through agents and APIs, so answers arrive where people actually work.
Together these turn a general model into a private assistant grounded in your material. You can read more about the engineering in how Od1n is built.
Doesn't on-premise AI need a server room?
No. Because Od1n is a small 3B model, it does not require a data centre or specialised infrastructure. The architecture is designed to run on ordinary hardware – a standard server with a GPU – which most organisations already have or can acquire easily.
For a single team or a proof of concept, modest hardware is enough. For production use with many concurrent users, a small dedicated GPU card improves response times and throughput. Either way, you are not committing to the cost and complexity that large frontier models demand.
How long does adaptation take?
Adapting Od1n to your organisation is measured in hours, not weeks. That is because the main adaptation happens through RAG – pointing the system at your documents – rather than retraining a separate model for every customer.
The base model is refined once through fine-tuning, and then each deployment is tailored by connecting your own knowledge base and integrations. Adding new documents or updating procedures does not require a new training run; you update the sources and the answers follow. This keeps deployment fast and keeps the system current as your material changes.
Who is on-premise AI for?
On-premise or EEA-hosted AI matters most where information is sensitive, confidential or regulated. Od1n's architecture is a strong fit for:
- Public sector – where citizen data and case files cannot be sent to foreign clouds.
- Healthcare – patient information and clinical records that demand strict control.
- Legal – privileged documents, contracts and case material.
- Finance – confidential client and transaction data under regulatory scrutiny.
- Companies with EU or Norway operations that hold sensitive or confidential information and want to keep it within their own boundaries.
If digital sovereignty is a strategic concern for your organisation, our perspective on digital sovereignty explains why control over your own AI stack is becoming a competitive and compliance issue rather than a nice-to-have.
Why does data staying in the building matter?
When you use a global AI service, your prompts and documents leave your control and fall under another provider's terms and another country's laws. For confidential material, that is often unacceptable – legally, contractually, or simply on principle.
With an on-premise or EEA-hosted deployment, the data path is short and auditable: input goes to the model on your infrastructure, answers come back, and nothing is shared externally. You keep ownership, and you keep the audit trail.
How do you get started?
Getting started begins with a needs conversation. We map what your organisation wants the model to do, which documents and systems it should draw on, and what your infrastructure and compliance requirements are. From there, Od1n is set up on your own infrastructure – hosted in the EEA or fully on-premise – and connected to your knowledge base and tools.
Od1n is a product of EZ-Fix AS in Oslo. To explore a deployment, see Od1n for business or contact us directly at post@ez-fix.no or +47 977 33 439.