AI without the Cloud – Fully Self-Hosted

Many companies have concerns about using commercial AI services:

• What happens to the data?
• Is everything properly secured?
• Is it legally or policy-wise permitted to send data to such services?
• Can we continue using this service even if the geopolitical situation worsens?

Fortunately, there is a fast and relatively inexpensive solution. It's very feasible to run AI locally, similar to ChatGPT, Gemini, or Copilot. Many open-source solutions are available, along with open/free AI models such as Llama, Gemini, DeepSeek, or Qwen.

However, setting this up can be challenging. Your team might not have the necessary expertise in-house, or if they do, they might simply be too busy.

I offer organizations a fully locally hosted AI solution, with no dependency on commercial cloud services. The system runs on your own hardware, is secure, extensible, and designed for full control over data and models.

This solution is ideal for:

• Biotech or medtech companies with sensitive data
• Research institutions that want to use AI without the cloud
• Teams looking to experiment with AI in a safe, isolated environment
• Organizations that value privacy, autonomy, and cost control

Since this is new technology, we start with a pilot setup. This allows you to quickly and safely discover where AI can add real value in your organization. If the pilot is successful, it can be expanded for use in critical processes.

Comparison: Commercial vs. Self-Hosted AI

Pilot Setup

The pilot setup consists of one server with an Nvidia RTX 5090 card. Mistral 7B runs on it (or another model in consultation). The AI is accessible via a web interface similar to ChatGPT, or via an OpenAI-compatible API for easy integration with other software.

You order the hardware yourself. Expect a cost of around €10,000–15,000. I handle installation, delivery, training, and documentation. After that, you can operate the system independently and with complete data security.

Price estimate: €10,000 excl. VAT

A total cost of €25,000 may seem significant, but for 40 employees over 3 years, this works out to just €17 per month per person — with no surprise costs down the road.

Production Setup

If the pilot setup is successful, it can be expanded with one or more production servers.

You order the hardware yourself, just like with the pilot. I deliver the systems fully operational, including training and documentation. I place the servers behind a load balancer so that the system continues to run even if one server goes down. This also allows for maintenance without downtime.

Price estimate: €20,000 excl. VAT

Additional Support

If you need extra help — for example, with specific AI needs or integration with your own software — I can assist. I’m an experienced software developer and can help review or contribute to your code as needed.

Rate: €150/hour excl. VAT

The Process

If you're interested, we’ll go through the following process. The goal is to be fully operational within 6 weeks:

1. We have an introductory meeting to assess whether this solution fits your needs.
2. I provide a proposal, and you approve it.
3. I send you the server specifications. You order and install the hardware on-site.
4. I configure the server for your intended use. This includes:
   1. Installing the software
   2. Configuring the chosen AI model
   3. Testing together on your systems
5. I then provide a full handover to your IT team. This includes a short training for maintenance and clear documentation.
6. Congratulations — you're now running a secure, powerful AI system independently, with no external dependencies.

Curious if this would work for you? Get in touch for a free introductory meeting. Email me at info@krolinventions.nl or see the Contact page.