Memory-Isolated, History-Preserved, and Future-Integrable Conversations: A User-Controlled Memory Architecture for Generative Artificial Intelligence

The development of generative artificial intelligence (AI) has increasingly incorporated persistent memory, cross-conversation personalization, and conversational history retrieval to support continuity between users and intelligent systems. Although these capabilities improve contextual relevance and reduce repetitive information exchange, they introduce an important architectural limitation: the context used to initiate a new conversation is often coupled with information accumulated during previous interactions. Consequently, users may have limited control over whether a particular conversation begins with previously established assumptions, preferences, personal information, or contextual dependencies. Existing temporary conversation mechanisms partially address this problem by providing reduced or disabled personalization, but they generally distinguish temporary interactions from persistent conversational history. Project-based memory isolation provides another partial solution, although its boundaries are typically determined by project membership rather than an independent control applied to an individual conversation. This paper proposes Memory-Isolated, History-Preserved, and Future-Integrable Conversations (MIHFI), a conceptual architecture for independently controlling the retrieval, preservation, and future reuse of conversational information in generative AI systems. The central principle is that a conversation should be capable of beginning without inheriting previously accumulated personal memory while remaining a permanent, retrievable conversation record that can subsequently contribute to future memory. Unlike conventional temporary conversations, MIHFI separates the initial context of an interaction from its persistence in conversational history and its potential contribution to subsequent personalization. The framework introduces three principal architectural dimensions: Memory Isolation, History Preservation, and Future Memory Integration. It further establishes a distinction between conversational memory retrieval, conversational record preservation, and memory-source eligibility, enabling these operations to be independently governed by user-defined controls. The proposed architecture is intended to support independent reasoning, exploratory research, creative experimentation, unbiased problem formulation, and user-directed knowledge development. A formal state model, operational workflow, design principles, and experimentally testable hypotheses are presented. Potential evaluation dimensions include memory contamination, contextual independence, retrieval accuracy, user autonomy, perceived privacy, conversational continuity, and cognitive diversity. The framework does not claim experimentally demonstrated performance improvements or universal novelty across all existing AI implementations. Instead, it establishes a testable architectural direction for advancing user-controlled memory management in generative AI. The central proposition is that conversational independence and long-term conversational continuity should not be treated as mutually exclusive properties.