JAKOVA-NET: EXAWATTS v1.1 – A Lightweight Work-Class and Power-Routing Layer for Personal Multi-AI Operating Systems

Personal AI operating systems are prone to the same failure mode as human projects: unbounded scope. Once multiple agents, nodes and pipelines are online, it becomes easy to spawn arbitrarily large tasks, leak time across devices, and accidentally turn a quick idea into an all-day refactor. EXAWATTS is a small tagging scheme designed to prevent that. It attaches two simple labels to every non-trivial unit of work: (i) a coarse size class (exa_class) from NANO to EXA, and (ii) a routing domain (domain) describing which part of the wider system the work belongs to (local machine, family rigs, cloud, or future campaigns). These tags are propagated into loop registries and overseer dashboards so humans can see—and cap—workload growth in real time. This paper formalises EXAWATTS v1.1 as used inside JAKOVA-NET, defines the class hierarchy and domains, describes integration with META_LOOP / XLOOP_MAP and JXJ pipelines, and gives concrete usage patterns for daily operation.