Local AI

Running AI on equipment that the user or their company controls, so prompts, files and answers stay there. On a phone or a computer it is called on-device AI. On-premises or self-hosted AI runs on a company's own servers, and edge AI on machines close to where the data comes from. To run a model locally, a user needs a copy of its weights and enough memory to hold them. If someone else made the model, the user gets that copy only if the maker publishes or licenses the weights. The largest models need several powerful machines working together, so permission to run a model and the hardware to run it are separate matters. Running a model locally replaces per-use fees with the cost of buying and operating the hardware. Many AI apps installed on a laptop still send prompts to remote servers. The app on the screen and the AI behind it then sit in different places.