Local models
AI on your own hardware, under your control
For data that must not go into anyone else’s cloud, open language models run on your own server or with a Swiss hosting provider of your choice. I choose a model that fits the task, set it up and measure whether it is good enough.
Three questions almost every company asks
- Open-weight models such as Mistral, Qwen, Llama or Gemma, with the licence checked first
- Runs on your server, on a machine with a graphics card or with a Swiss hosting provider
- An honest comparison with cloud models before you buy any hardware
I run local models on my own hardware, so I know where the limit is: nothing local comes close to the large cloud models. For narrowly defined tasks with sensitive data, though, it is often enough.
How we start
We compare an open model on your own hardware with a cloud model, using your own data. Only then do you decide whether the hardware is worth buying.