We design and build generative AI that earns its place in production — assistants and copilots, content and document generation, and AI agents — grounded in your own data and processes. A Dutch generative-AI partner that ships working systems, not slideware. EU AI Act- and GDPR-ready.
Last updated: 11 June 2026
Generative AI moved from novelty to a real productivity layer fast — drafting content, answering questions from your knowledge, reading and routing documents, and increasingly acting as agents that complete multi-step tasks. The hard part isn't the demo; it's making it accurate, safe and integrated into how your business actually runs.
We start with where generative AI genuinely pays off for you, build a working proof on your own data, then take it to production with the guardrails, evaluation and integration that separate a reliable tool from an impressive toy. Making the underlying model accurate and grounded is our LLM Optimisation work; this service is about designing and building the generative-AI solution around it.
Internal copilots and customer assistants that answer from your own knowledge — accurate, sourced and on-brand.
Draft, localise and repurpose copy, product text and marketing assets from your templates and tone, with a human in the loop.
Read, summarise and structure contracts, invoices, reports and forms — turning unstructured documents into usable data.
Agents that complete multi-step tasks across your tools — with the right guardrails on what they can and can't do.
Retrieval-augmented answers grounded in your documents, so responses stay current, sourced and trustworthy.
Generate and analyse images and mixed media where it adds value — product visuals, inspection, creative variations.
We find where generative AI genuinely pays off and the data and systems it touches — fixed-price.
By the second call you get a working proof on your own data — not a spec.
We integrate it into your tools with evaluation, guardrails and monitoring built in.
We track accuracy, cost and usage so the system keeps earning trust.
The market for generative AI development splits into two extremes, and most Dutch companies feel stuck between them. On one side sit the large enterprise consultancies — Xebia, Xomnia, Capgemini — whose minimum engagements and layered teams price out the average MKB and rarely leave you owning what was built. On the other side sit web and marketing agencies that rebranded as "AI" over a weekend and resell a thin wrapper around someone else's API. Crux Digits is a boutique, senior-led generative AI agency in the Netherlands that sits deliberately between those poles: small enough to keep the founder and senior engineers on your project, serious enough to ship GenAI systems that survive contact with production, EU AI Act and AVG audits.
Founded in 2022 and based in Nieuwegein in the province of Utrecht, we serve the Utrecht region and the wider Netherlands and Europe. We have delivered 13 case studies across healthcare, computer vision, natural-language processing and forecasting (client names stay confidential), so our GenAI consulting is grounded in shipped work rather than slide decks. When we engage, senior people stay on the build from the first audit to the production handover — and the deliberate end state is that you, not us, own the running solution.
Owning the solution matters more in generative AI than in almost any other engineering discipline, because the cost and behaviour of these systems shift weekly as models, prices and regulation move. If a vendor keeps the prompts, the retrieval pipeline and the evaluation harness locked in their own account, you are renting your own product. We hand over the code, the configuration and the documentation, and we explain the trade-offs in plain language so your team can run, extend and audit the system after we step back.
The hero of this page covers the use cases — assistants, copilots, content and document generation, AI agents, retrieval-augmented search. Below the surface, generative AI development is a stack of engineering decisions that determine whether a pilot becomes a dependable tool or quietly gets switched off. Most of the value, and most of the risk, lives in the layers users never see.
Retrieval-augmented generation (RAG) is the technique that makes AI assistants and AI copilots answer from your documents rather than from the model's generic training data. In practice that means ingesting your knowledge base, splitting it sensibly, embedding it, retrieving the right passages at query time and feeding them to the model with the citations attached. Done badly, RAG retrieves irrelevant chunks and the assistant confidently invents answers. Done well, every response is sourced and checkable. We treat retrieval quality as an engineering problem with measurable accuracy, not a setting you toggle on. The deeper grounding, prompt and fine-tuning work lives under our LLM Optimisation service; this page is about designing and building the GenAI product around it.
An assistant answers; an AI agent acts. Agents plan multi-step tasks and call your tools — looking up a record, drafting a reply, updating a system, triggering a workflow — to complete work end to end. That power is exactly why scope control matters. We define precisely which actions an agent may take, which need human approval, and which are forbidden, then we log every step so the behaviour is auditable. For most Dutch SMEs the right first agent is narrow and bounded: a single workflow that removes a repetitive task, not an open-ended autonomous system. That discipline is also the through-line of our broader AI automation work.
There is no single correct model. We build on leading API models (OpenAI, Anthropic, Google) when their capability earns it, and we deploy open-weight models such as Llama or Mistral on infrastructure you control when privacy, data residency or cost demands it. The choice follows the use case and your constraints — never a vendor preference. For a Dutch healthcare or finance client handling sensitive records, an EU-hosted or self-hosted model can be the difference between a compliant deployment and one legal will not sign off.
Generative AI is often pitched as an enterprise-only capability. For the Dutch mid-market the opposite is true: the highest-return projects are usually small, concrete and unglamorous. The pattern that works for MKB companies is to pick one bounded use case, prove it on real data, and only then widen the scope.
Because we also act as the AI engineering partner behind marketing and web agencies, we see the same practical wins repeat across sectors. Crux is not itself a marketing or web agency — we are the GenAI build team those agencies bring in when a client needs real generative AI rather than a demo.
The EU AI Act and the GDPR/AVG shape what you are allowed to build and how. We design for both from day one: data minimisation, access control, clear records of what the system does and why, and EU-hosted or private models where the data calls for it. Building compliance in at the start is far cheaper than retrofitting it onto a system already in production — and for regulated Dutch sectors it is the only way a generative AI project reaches launch at all.
Generative AI pricing is usually opaque, which makes budgeting impossible. We work in transparent, fixed steps so you know the commitment before each one (all figures exclude VAT):
You can stop after any step. Starting with the audit means a few thousand euros buys clarity on whether a use case is worth building before you spend tens of thousands building it. See the full breakdown on our pricing page.
The four-step rhythm on this page — audit, MVP, deploy, improve — describes the shape of every engagement, but a few details matter for setting expectations. The audit is fixed-price and fast, so you are never paying open-ended discovery fees. The proof of concept lands in weeks rather than months, and it runs on your own data so the output is honest rather than a curated demo. At the production stage, evaluation and monitoring are built in from the start: we track accuracy, cost per query and real usage, because a generative AI system that is not measured is one that quietly drifts.
If you want the strategy view before the build view, our AI consulting page sets out how we frame an AI roadmap, and our case studies show the kind of NLP, computer-vision and forecasting work we have delivered. To understand who stays on your project and why senior involvement is non-negotiable for us, read more about Crux Digits. When you are ready, tell us the one task or workflow you most want to fix — we will scope a focused generative AI proof of concept on your own data in a free consultation, and you will know the cost and the plan before we write a line of code.
Generative AI (this service) is designing and building the solution — the assistant, agent, content or document workflow — around a model. LLM Optimisation is the layer that makes the underlying model accurate, grounded and affordable (RAG, prompts, fine-tuning, evaluation). Most projects use both; we do them under one roof.
Usually we build on leading API models (OpenAI, Anthropic, Google, or open-source like Llama/Mistral) and ground them in your data. Where privacy or cost demands it, we deploy open models you host. We pick the model to fit your use case and constraints, not the other way round.
We ground responses in your own knowledge (RAG), constrain them with system design and guardrails, evaluate against real examples, and keep a human in the loop for anything high-stakes — so output is accurate, sourced and consistent.
Yes. We design for GDPR and the EU AI Act from day one — data minimisation, access control, EU-hosted or private models where needed — so your data stays protected and you stay compliant.
A free 30-minute consultation is usually within days; a working proof of concept on your own data typically lands in weeks, not months.
We work in fixed steps: an AI audit & strategy at €2,500, a proof of concept at €20,000 and a production launch from €50,000 (excl. VAT). You know where you stand before each step.
Especially so. Most SME wins are practical — drafting, document handling, customer answers — and starting with one bounded use case keeps the risk low and the payback fast.
Yes — we connect to your CRM, content systems, document stores and internal tools through their APIs, so the AI works inside your existing workflow rather than as a separate app.
Tell us what you want to build or automate — we'll scope a focused proof on your own data in a free consultation.
Book a free consultation →