As powerful open models become portable and private, neighborhoods begin keeping sovereign AIs as local memory keepers for emergencies, repairs, and disputed facts.
What starts as a backup tool becomes a parallel civic institution. Tenant associations, fishing cooperatives, and mutual-aid groups train local models on evacuation routes, machine manuals, dialects, land records, and hard-learned practical knowledge. In normal times these systems help with repairs and translation; in crises they become trusted guides when cloud services fail or official channels lag. Over time, communities rely less on universal platforms and more on locally governed intelligence, creating a patchwork knowledge order that is resilient, intimate, and difficult for any single authority to fully supervise.
At 6:40 a.m. in a flooded apartment tower in Busan, a building manager named Hyejin unlocks the basement battery room and asks the local model which elevator shaft can be safely reopened before school commute hour.
The same autonomy that preserves local knowledge can also preserve local delusion. A neighborhood model can become a fortress for rumor, sectarian memory, or technically confident bad advice, and appeals to outside expertise may arrive too late to matter.