Memory Architecture in S-AI-Gpt: From Contextual Adaptation to Hormonal Modulation

This article presents a biologically inspired memory architecture embedded within the Sparse Artificial Intelligence – Generative Pretrained Transformer (S-AI-GPT) conversational framework. Addressing the limitations of stateless Large Language Models (LLMs), the system integrates three complementary components: a Dynamic Contextual Memory (DCM) for short-term working retention, a GPTMemoryAgent for long-term personalized storage, and a GPT-MemoryGland for affective trace encoding and modulation. These components are orchestrated by a hormonal engine, enabling adaptive forgetting, emotional persistence, and context-aware prioritization of memory traces. Unlike typical passive memory modules, this architecture introduces an active, symbolic, and controllable memory system: memory traces can trigger internal hormonal signals, are stored in a structured and interpretable form, and can be selectively reinforced, inhibited, or reorganized by the GPT-MetaAgent. The proposed model provides a promising foundation for building frugal, adaptive, and explainable lifelong memory systems in conversational AI.