Oriented Context AI: A Human-Centered Model for Structured and Transparent Dialogue

This note presents the concept of Oriented Context AI, a human-centered model designed to improve coherence and transparency in long-form conversations between humans and large language models (LLMs).While current AI systems such as ChatGPT or Claude rely on hidden, algorithmic context memory, users often lose visibility and control over what the model “remembers” or prioritizes. This results in conversational drift and cognitive overload during extended dialogue. To address this, the Oriented Context AI model introduces a three-column interactive interface composed of: Prompt / Response – the conversational flow; Key Elements – an automatically generated list summarizing the essential ideas of each exchange; Root Memory (Persistent) – a user-editable area where key elements can be dragged, anchored, or removed at any time. This structure transforms the conversation into a shared cognitive space where users can explicitly manage the system’s long-term context.The resulting workflow — extract → validate → anchor — balances human agency and AI continuity, offering a transparent and explainable alternative to conventional hidden-memory chat systems. The document includes the full LaTeX source, figure, and bibliography illustrating how this design could be implemented in practice.It is intended as an open contribution for developers, UX researchers, and AI labs exploring more transparent and controllable conversational interfaces.