Local AI

Running capable AI on hardware you control is newly practical: open-weight models, cheap inference stacks and consumer GPUs put private, offline AI within reach of councils, firms, schools and households. The UK conversation still defaults to renting US frontier APIs, so the questions that matter here go unasked: what should run locally, who maintains the reference deployments, and what national capability exists if the rented option is withdrawn or repriced. This domain collects the gaps in that local capability. It starts small, seeded from the AI-crisis dossier, and grows as entries are suggested and verified.