Off-grid personal compute
Local, self-improving, off the grid — measured against today's silicon.
Define the smallest self-sustaining personal compute stack that can run useful local AI without constant cloud dependence.
Problem
Pockot
Pockot
Useful local AI is no longer only a datacenter question. Pockot studies the constrained version: a pocket-scale or field-portable compute stack with a real power budget, local storage, model limits, and a clear fallback mode when networks disappear.
The work stays measured against current silicon, batteries, and open model runtimes. No apocalypse theater; just the bill of materials, watt-hours, and capability gaps.
Operating Rule
Receipts before claims.
Every public fact, number, or model baseline is attached to a visible source. Estimates are marked as estimates. Boundaries stay close to the result.
Reference Battery100 WhFAA boundary
Small Models1B / 3Bedge baseline
BoundaryNo guaranteeheuristic model