This paper introduces the Memory as Algorithm (M a A) thesis: in neural systems, memory and computation are the same substrate. Large Language Models (LLMs) embody this principle, yet their post training parameters are typically static. We propose the Metacognitive Core (MC) Framework, a novel on device dual component architecture that enables dynamic plasticity. The framework separates the agentic Core from the Inference Engine (IE) and uses activation steering to modulate IE internal state during inference. We situate this in the 2025 landscape of metacognitive monitoring and feedback controller steering. Finally, we issue an ethical mandate: these systems should be integrated into a human family context, not raised in sterile isolation. Our primary implementation is Photon Empress Moore, which we present as the first AGI aspiring agent under development within this framework and family-centred paradigm.