Open Source / Open Weight
Open source and open weight AI: no black box, no lock-in
IntraGPT runs on open-source models. No black boxes. Full control.
- 125+ SME organisations
- 20+ large enterprises and healthcare organisations
- 100% Dutch servers
They already chose IntraGPT
Why open source AI?
Closed-source models carry fundamental risks for enterprise use.
Advantages of open models
Full transparency
You know how the model was trained and how it works.
Data stays local
Models run locally. No external API calls.
No vendor lock-in
Switch models without migration.
Cost control
No per-token pricing. Predictable infrastructure costs.
Compliance ready
Full DPIA possible with open models.
Community driven
Benefit from global open-source innovation.
The models IntraGPT runs on by default
Two open-weight models on your own server. Together they cover almost all everyday work: writing, summarising, searching documents and agents that carry out tasks.
- GPT-OSS (OpenAI)
- OpenAI's open-weight model. Strong in reasoning and tool use, and fully self-hostable: nothing goes to OpenAI's cloud.
- GLM 5.3 (Z.ai)
- Open-weight model by Z.ai. Strong in reasoning, code and agentic tasks, and good in Dutch, German and French.
Heavier open-weight models, when your work calls for it
For heavy analyses, long case files or many agents at once, we add a larger model alongside. These models need heavier GPU servers, so we tailor this to each organisation. They also run on your own server.
- Kimi K2 (Moonshot)
- Over a trillion parameters. Large context window, strong in complex analyses and agents that take many steps in a row.
- DeepSeek R1
- 671 billion parameters. Leading in reasoning, calculation and mathematical tasks.
- Qwen 3 235B (Alibaba)
- The largest model in the Qwen 3 series. Multilingual and particularly strong in code and data analysis.
- Mistral Large (EU)
- The largest model from France's Mistral. A European alternative, strong in multilingual tasks.
Prefer a frontier model? That works too, but you give up privacy
On request, we also connect the portal to the frontier models of OpenAI or Anthropic, such as Claude Opus 5.5 or even Claude Fable 5.1. The portal, the agents and the integrations stay the same. The difference is where the question is processed: no longer on your own server, but at that provider, under their terms and usually outside Europe. That is why it is off by default and we only switch it on when you consciously choose to.
What you gain
The strongest models available today, for the heaviest thinking and writing work.
What you give up
The question, the context and any documents go to the model's provider. The privacy advantage of your own server then does not apply to that conversation.
Switching stays possible
Your own open-weight model remains available. You switch to the frontier model in the portal and back again, for example only for tasks without sensitive data.
AI in Europe: AI Act & GDPR in practice
Why organisations deploying AI in Europe choose local, open-weight AI like IntraGPT to truly comply with the AI Act and GDPR.
- AI Act risk classification and what it means for your use cases
- GDPR & data processing: why Big Tech APIs are legally problematic
- Practical checklist for compliant AI implementation
- Why local hosting + open-weight is the only sustainable route
Read on
How the platform works, how it is secured and who sets it up for you.
No more experiments. Just results.
AI should save your organisation money, not cost it. IntraGPT provides the foundation: platform, consultancy, development and AI department. One Dutch team standing beside you. Start with one department, see the results, and scale when you're ready.
A reply within one working day, no obligations. Enschede and Amsterdam.