Artificial intelligence deployments at the enterprise edge face a problem that centralized AI architectures are structurally unequipped to solve. Current systems are built on four external dependencies: persistent network connectivity, centralized compute, ephemeral context, and continuous human orchestration through a chat interface. In environments where connectivity is denied, degraded, intermittent, or limited (DDIL), these dependencies do not degrade gracefully. They fail completely. This paper defines a shift in how AI systems are architected for deployment in mission-critical, connectivity-constrained environments. The central concept is the self-sufficient cognitive system: a bounded, deployable cognitive node capable of operating when the link to the enterprise is severed. Self-sufficient is a precise term — it does not mean autonomous; it means independent of persistent internet connectivity. The node loses its tether to the cloud. It does not lose its tether to the human. The architecture introduced here — Alistair Prime in a Box — embeds five capabilities within a single deployable unit: local inference, persistent memory, agentic orchestration, governance and policy enforcement, and execution capability. The core contribution is a structured dependency decomposition model that systematically identifies, categorizes, and eliminates or localizes every external dependency a cognitive system carries, producing a DDIL-tolerant architecture with explicit degradation tiers that preserve function — and governance — as connectivity erodes. Rights envelope: Citation permitted with full attribution. No reproduction, redistribution, or derivative works without written permission. AI/ML training use disallowed. See the citation policy at https://nonsequitur.tech/pubs/citation-policy/ for the full rights envelope. Canonical site URL: https://nonsequitur.tech/white-papers/alistair-prime-in-a-box/ Public archive: yks-pubs/papers/alistair-prime-in-a-box-v1-preprint.pdf