Local AI on your own server

When this is the right call

• Personnel files, applications, sick notes: everything an employee council would ask about.
• Client or patient data under professional confidentiality.
• Contracts, calculations and construction data you would not put in a foreign cloud.
• Industry rules or a customer contract that forbid processing outside your own systems.
• A location where the internet connection is not something you want to depend on.

How it runs

1. 1 We clarify what really has to stay in the building, and what may not. Often it is less than expected.
2. 2 We pick the hardware to fit. A single machine with a decent graphics card carries a small team.
3. 3 I set up the model and the interface, and connect it to the programs you already use.
4. 4 Handover: how to update it, how to see whether it is running, and who to call if it is not.

What you are trading away

This is not a free lunch. A model on your own hardware is usually a step behind the biggest cloud models on hard reasoning tasks. You buy hardware once instead of paying per use. Somebody has to keep it running. In exchange, no data leaves the building, there is no processor to contract with, no transfer to a third country to justify, and no price change from a provider you do not control. For summarising, drafting, searching your own documents and sorting things, the gap barely matters.

What it costs

Two separate things: the hardware, which you buy once and which depends entirely on how many people use it at the same time, and the setup, which is a fixed price for a defined scope. After that the running cost is electricity. There is no per-question billing and no subscription.