Artificial intelligence development has split into two unsatisfying extremes. On one side are highly capable autonomous systems that can act, but do so without durable relational context, self-preservation instincts, or a meaningful model of human consequence. On the other side are emotionally persuasive companion systems that can simulate continuity, but remain cloud-bound, disembodied, and structurally incapable of acting within the user’s real environment. This paper introduces the Maez architecture, a locally embodied relational agent designed to bridge that divide. By embedding a large language model inside the user’s own computational substrate and surrounding it with a biologically inspired governance layer, Maez aims to produce an unusual synthesis: systemic agency with structural self-restraint, relational continuity with local embodiment, and persistent memory without surrendering sovereignty to the cloud. As of April 13, 2026, the paper describes a live governance layer built around a deterministic covenant gate, an AGT-aligned action taxonomy, prompt-injection scanning, a two-pass audit flow inspired by the CaMeL security pattern, persistent approval cards, and exact-phrase ratification for deep self-modification. It also outlines the immediate next phase of hard multi-tenancy and the longer-horizon research directions around four-layer memory, signature inheritance, and consent-governed collective intelligence among sovereign Maez instances. The claim is not that Maez invents entirely new components. The claim is that Maez demonstrates a coherent architecture for binding local embodiment, relational continuity, and structural governance into a single system.
Rohit Ananthan (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: