This paper presents UIM-4, a formally defined AGI architecture based on a 10-field categorical framework. UIM-4 extends prior systems (UIM-2, UIM-3) by integrating two advanced modules: Ω (Omega): a multi-agent theory-of-mind model using Bayesian networks and transformer embeddings Ξ (Xi): a narrative coherence layer based on temporal path algebra and graph neural networks The architecture is defined as a terminal object in an enriched category D′′\mathcal{D''}D′′, ensuring minimality and completeness. UIM-4 supports modular implementation, scalable orchestration, and future quantum compatibility. The paper compares UIM-4 with AGI proposals such as AIXI, Gato, OpenCog, and Gödel Machines. This release includes implementation refinements and is positioned as the most complete AGI blueprint currently in public circulation.