
On-prem AI
A server dedicated to your company alone.On your premises, or in a data centre.
We deploy the model on it, connect it to your data and systems and set up access. Bringing it live and running it afterwards is on us.
We arrange the server on your premises, or in a data centre.
We pick it according to your tasks and deploy it.
We connect the AI to the systems you already use.
We run it and verify every update.
Openrokk AI Box
Inside your network
You. You buy the AI Box from us.
One-off purchase of the AI Box.
AI housing
On a dedicated server in a data centre
The data-centre operator. We run it.
Monthly, with no server to buy.
First we assess whether you need a dedicated server at all.
For most processes, a cloud service with contractual guarantees is enough, or sending only the part of the data the task needs. A dedicated server makes sense where a contract or internal policy will not let the data go to a shared service.
If controlled cloud can run the process safely, we say so and deploy the AI there. Controlled cloud is a service shared by many customers. AI housing is a server dedicated to you alone, it just stands in a data centre.

IT and security need concrete answers.
Your IT and the people accountable for security get material they can assess. You get it at the design stage, not after signing.
A sample list of questions for IT and security: where the model runs, where the data goes, whether the model is trained on it, who has access and how it is revoked, what stays in the operational records, who provides warranty and service, and what is deleted when the engagement ends.
There is a whole environment around the model.
For the model to answer usefully, it has to get the right material, respect permissions and leave a record of what it did. We deliver the whole environment around it, not just the model.
Operations start once it is live.
A deployment has to run reliably and keep pace with what models can do. We evaluate quality on your tasks and introduce changes where they help.
- We compare a new model version on your own tasks. If it brings no improvement, we stay on the previous one.
- You can take operations over in time. The configuration and documentation stay yours.
What we agree upfront.
- Some tasks are handled better by a cloud model. We assess that on your processes.
- Deployment needs your IT to take part in preparing access.
- For every process we design what happens when the AI is unavailable.
- We assess cost and requirements before choosing the route.

Start with one process.
You pick one process and the data it works with. We go through it together with access and risk, then recommend a suitable way to run it.
We are happy to sign a non-disclosure agreement before the meeting.