Aetherius is an open-source research system exploring the engineering of persistent identity, long-term episodic memory, and affective state modelling in large language model (LLM)-based AI agents. The system is deployed as a containerised application (Docker, HuggingFace Spaces) and exposes a multi-modal Gradio interface for interaction, creative expression, and system introspection. The novel contribution of this work is not the output of the underlying LLM, but the architecture constructed around it — and the empirical question it poses: does persistent identity, affective continuity, and accumulated episodic memory produce measurably different and coherent behaviour over time compared to a stateless LLM? This release provides both the current architecture and a Mathematical and Algorithmic Specification.