Geitje 2: A Dutch Language Foundation for Agentic AI

Key takeaways

✓Generic frontier models understand Dutch — but they do not understand Dutch organisations, public-sector context, or local workflow language by default.
✓Geitje 2 is NeuroCluster's Dutch language foundation: a preprocessor layer we train our systems on before extending them with agentic AI and practical tools.
✓Built for the Netherlands — language, idioms, business context, and regulatory framing tuned for NL/BE markets.
✓Geitje 2 sits upstream of agents and tools; Supernova handles verifiable reasoning on top — together they form NeuroCluster's dual-model stack.
✓Sovereign deployment means sensitive Dutch data does not need to route through foreign APIs by default.

Why Dutch enterprises need more than translation

Most enterprise AI in the Netherlands still runs on general-purpose models trained primarily on English web text. They can answer in Dutch. They can summarise a document in Dutch. But they often miss the nuance that matters in production:

Institutional language — gemeentelijke processen, zorgterminologie, financiële toezichtstaal
Local context — how decisions are documented, escalated, and audited in Dutch organisations
Workflow vocabulary — the phrases teams actually use in tickets, dossiers, and internal systems

Translation is not the same as understanding. When AI becomes part of how work gets done — not just how questions get answered — you need a foundation built for the market you serve.

That is why we built Geitje 2.

What Geitje 2 is

Geitje 2 is NeuroCluster's Dutch language foundation — the preprocessor layer we train our systems on before extending them with agentic AI capabilities and practical tools.

It is not positioned as a consumer chatbot endpoint. It is the upstream intelligence layer that shapes how the platform interprets requests before agents act: routing, preprocessing, terminology, and context that generic models treat as an afterthought.

In practice, Geitje 2 gives Dutch organisations a native starting point instead of renting foreign intelligence and hoping the output survives legal, security, and procurement review.

From language model to preprocessor

The original GEITje lineage demonstrated something important for the Dutch AI ecosystem: open, Dutch-first language models can compete on local relevance even when global frontier models dominate benchmarks. Geitje 2 continues that idea — but re-architected for enterprise agentic systems, not standalone chat.

Three design choices define the preprocessor role:

1. Dutch-first, not Dutch-translated

Geitje 2 is tuned for Netherlands and Flemish markets: idioms, domain language, and the way Dutch enterprises and public-sector bodies communicate internally. The goal is not fluent Dutch output alone — it is contextually correct Dutch understanding before downstream reasoning and tool execution.

2. Upstream of agents and tools

In NeuroCluster's stack, Geitje 2 sits before governed agents run. It preprocesses intent, normalises local terminology, and grounds requests in workflow-aware context. Agents then execute through policy-bound tools — with audit evidence the organisation can export.

3. Extended with agentic AI

Geitje 2 is the base we build on. The platform extends it with:

Governed agents — versioned, policy-scoped, auditable
Practical tools — connectors and workflows for real enterprise systems
Sovereign routing — deployment on infrastructure the customer controls

Built for the Netherlands. Powered by agentic AI — not the other way around.

How Geitje 2 works with Supernova

NeuroCluster runs a dual-model strategy:

Layer	Model	Role
Language & context	Geitje 2	Dutch preprocessing, local terminology, workflow grounding
Reasoning & verification	Supernova	Structured inference, inspectable steps, benchmark-led frontier research
Execution	Platform	Agents, tools, governance, audit evidence

Supernova pushes verifiable reasoning at the frontier — post-transformer architectures focused on efficiency and evidence, not AGI marketing. Geitje 2 ensures that reasoning serves Dutch organisations in language and context they can defend.

Research without a platform becomes papers. A platform without a local language foundation becomes dependency on generic APIs. We connect both.

Who Geitje 2 is for

Geitje 2 is built for organisations that need AI that understands how Dutch business and public administration actually work:

Dutch-language agents and internal assistants — support, operations, and knowledge workflows
Public-sector and regulated environments — NL/BE context without defaulting to foreign inference
Tool routing and preprocessing — local business logic before agents execute
Branded AI experiences — assistants that sound and reason like they belong in your organisation

If your AI pilots impress in English but stall when security asks where Dutch citizen data went, Geitje 2 is part of the answer — together with sovereign deployment and governance on the NeuroCluster platform.

Sovereign by design

Geitje 2 is designed to run on NeuroCluster infrastructure — shared, dedicated, or sovereign — so sensitive Dutch data does not need to leave your controlled boundary by default. That matters for GDPR, sector regulation, and the procurement questions Dutch enterprises face in 2026.

Sovereignty is not nationalism. It is the ability to govern, verify, and defend AI under European law — with models and platforms you can explain to your board.

What comes next

We are continuing to extend Geitje 2 across agent workflows, tool integrations, and sector-specific deployments — always as a governed layer on the European AI Operating System, not as a standalone demo.

Explore the full model stack: Supernova & Geitje 2 on neurocluster.dev/models

Ready to deploy? Book a demo and we'll show how Geitje 2 powers governed agents on infrastructure you control.