Skip to main content
ZIVOS
Status: In development

Norwegian AI • Built for reality

We are building Norwegian intelligence.

We are developing Arktis AI, a Norwegian AI model for the language, domains and workflows that general models do not always understand well enough — and for organisations that require precision, control and trust.

Request early accessExplore the technology
  • Developed in Norway
  • Norwegian language understanding
  • Built for enterprises
  • Flexible deployment
  • Responsible AI
  • European jurisdiction

The company is in its development phase. Capabilities that are not yet documented are not published on this site.

Global AI understands a lot. Norwegian reality demands more.

Most organisations do not meet AI as a general question, but as a concrete workflow: a document to assess, a case to prepare, a regulation to comply with.

  • Norwegian language and domain-specific terminology.
  • Internal documents that were never part of public training data.
  • Sensitive information requiring a defined legal basis.
  • Regulatory requirements that must be documented afterwards.
  • The need to trace an answer back to its source.
  • Results that must be verifiable, not merely convincing.

General model

  • Broad knowledge across languages and domains.
  • Trained on publicly available material.
  • Often operated outside European jurisdiction.
  • Limited insight into data sources and changes.

Specialised model

  • Adapted to Norwegian professional language and document structure.
  • Evaluated against the tasks it is actually meant to solve.
  • Can run in your own or a European environment.
  • Documented behaviour, limitations and version history.

A specialised model is not automatically better at everything. The point is relevance, control and fit for the specific task.

In development

A model built for the language, the domains and the reality.

Arktis AI is being developed as infrastructure for digital work: conversational and agent-based interfaces that solve complex tasks, work with documents and carry out multi-step assignments under human control.

Read about the modelSee the platform
Model
Arktis AI 1.0
Status
In development
Developed in
Lierne, Norway
Languages in development
Norwegian, English, German, French, Spanish
Deployment
European cloud, private cloud, on-premise
Regulatory baseline
EU AI Act and GDPR

Layers of the work

  1. 01Data foundation

    Defining and documenting which sources are included, and which are deliberately excluded.

  2. 02Training and adaptation

    Adaptation to Norwegian language and terminology, with every change versioned.

  3. 03Domain specialisation

    Tuning towards the document types and workflows of the prioritised industries.

  4. 04Evaluation

    A framework describing method, datasets and limitations before any result is published.

  5. 05Safety mechanisms

    Input and output controls, access management and handling of sensitive content.

  6. 06Deployment

    Options from a managed API to installation inside the customer's own environment.

  7. 07Monitoring

    Logging and follow-up of model behaviour in production, without storing content unnecessarily.

Capabilities in development

Everything below is either under active development or on the roadmap. Nothing is presented as a finished, verified capability.

  • Document understanding

    Reading, interpreting and structuring long documents in a Norwegian professional context.
    Internal documents, contracts, reports and case files.
    Less preparation time before a professional assessment.
    Requires human review before the answer informs a decision.

  • Knowledge retrieval with citations

    Finding relevant passages across a closed document collection.
    The organisation's own access-controlled collections.
    Knowledge that already exists actually becomes available.
    Quality depends on the underlying material being current and correct.

  • Summarisation

    Condensing extensive material into a structured basis for decisions.
    Meeting minutes, case documents, technical reports.
    Faster overview without losing the trail back to the source.
    Summarisation removes nuance; the original is always authoritative.

  • Structured extraction

    Pulling defined fields out of unstructured text.
    Forms, invoices, claims, contracts.
    Manual entry replaced by verifiable completion.
    Uncertain fields must be flagged, never guessed.

  • Analysis and reasoning

    Breaking complex tasks into verifiable steps.
    Combinations of documents and structured data.
    Better input for complex assessments.
    Reasoning is not proof and must be checked.

  • Tool use and workflows

    On the roadmap
    Agent-based multi-step tasks with defined tools.
    Systems the organisation explicitly grants access to.
    Routine work carried out with traceable steps.
    Every action must be stoppable and auditable.

  • Code

    On the roadmap
    Writing and reviewing code as part of a workflow.
    Code bases the organisation explicitly grants access to.
    Faster development work with human review.
    All code must be reviewed before it reaches production.

  • Multimodal understanding

    On the roadmap
    Interpreting text together with tables, forms and scanned documents.
    Documents that are not available as plain text.
    Fewer manual steps in the document flow.
    Quality depends on the legibility of the source material.

Interactive product preview

What the workspace looks like

Below is a curated view of the interface we are building. It shows how an answer should be presented: with sources, control and clear model attribution.

Example tasks

  • Summarise this governance document with clear source references.
  • Identify deviations and ambiguous provisions in this agreement.
  • Turn this financial report into a basis for decision.
  • Find related points across these documents.
  • Classify and structure the information in this document.

This is a pre-defined preview. It is not connected to a live model, and the content is an illustrative example.

Task

Summarise this governance document with clear source references.

governance-document-2026.pdf

Answer

  • The document sets out three main priorities for the period, with assigned responsibilities.
  • Two measures have no stated deadline. These are flagged for follow-up rather than interpreted.
  • The budget section refers to an appendix that is not included in the submitted material.

Sources

  • Section 2.1 — Priorities
  • Section 4.3 — Measures
  • Section 6 — Budget

Control

This answer requires approval by the responsible case officer before use.

Interactive product preview

Your model. Your environment. Your requirements.

The deployment model is part of the product. Where the model runs determines which requirements it can meet.

Managed API

The simplest entry point for evaluation and development, operated by ZIVOS.

European cloud

Operation with European cloud providers inside European jurisdiction.

Private cloud

A dedicated environment for organisations with their own isolation requirements.

Your own environment (on-premise)

Installation inside your own infrastructure, including restricted networks.

Hybrid

A combination where sensitive processing stays local and the rest can run centrally.

  1. Source system
  2. Access control
  3. Model
  4. Log and traceability
  5. User

Specific regions and providers will be published only once agreements are in place and can be documented.

Control is a feature, not a footnote.

We build the EU AI Act and GDPR into the architecture from the first version, and document the choices as we go.

Visit the trust centre

  • Data and privacy

    Data minimisation, a defined legal basis and clear rules for what is stored and for how long.

  • Evaluation and quality

    An evaluation framework describing method and limitations before results are published.

  • Security and responsible use

    Access control, input and output checks, and clear guidance on acceptable use.

  • Transparency and documentation

    Model card, change log and documented limitations published together with the model.

Built for work that actually matters.

Four prioritised areas where linguistic precision and regulatory control are decisive.

See details

Claims are not documentation.

We publish no results before the method can withstand review.

The evaluation framework will be published together with the first model version.

Every published result will include

  • Date of testing
  • Model version
  • Dataset or method
  • Source
  • Known limitations
  • Comparison baseline

Research and news

Professional updates from the development work.

No content has been published yet. The first updates will follow the model work.

See all

Frequently asked questions

What is ZIVOS?
ZIVOS AS is a Norwegian technology company developing AI models and infrastructure for digital work. The company is based in Lierne.
What is Arktis AI 1.0?
Arktis AI 1.0 is the ZIVOS AI model for Norwegian language and professional context. It is in development and not yet generally available.
Who is the model built for?
Norwegian and European organisations with high requirements for precision, traceability and regulatory control — especially public sector, finance, legal and industry.
How does it differ from general AI models?
Arktis AI is developed for Norwegian professional language, document structure and workflows, with deployment inside European jurisdiction and documented limitations.
Where is the technology developed?
The technology is developed in Norway, based in Lierne.
How can organisations get access?
By requesting early access through the contact form. The model is in development, and access is granted in dialogue with each organisation.
Which languages are supported?
Development covers Norwegian, English, German, French and Spanish. Norwegian is the main priority.
How does the company work with safety and responsibility?
The EU AI Act and GDPR underpin the architecture. Model card, evaluation method and limitations are published together with the model.

Norwegian AI should not only be used. It should be built.

Talk to us about how Arktis AI can be developed, evaluated and deployed in your organisation.

Request early accessContact the team