Answers are grounded in this paper and cite their sources.
Generative artificial intelligence platforms increasingly support long-running conversations containing extensive textual exchanges, uploaded documents, images, generated media, tool interactions, and accumulated contextual information. As these conversations grow, their persistent data footprint and retrieval requirements may increase, making them difficult to open, transfer, render, or manage, particularly on devices with limited memory, processing capacity, or network resources. A conversation that remains accessible on a desktop computer may become difficult to load on a smartphone. Existing research has investigated prompt compression, conversational memory compression, semantic summarization, incremental dialogue compression, and the reduction of model inference context. However, these approaches do not necessarily provide users with direct control over the persistent compression of an existing conversation as an application-level object. This paper proposes User-Controlled Dual-Entry Persistent Conversation Compression for Generative AI (UCCPC-GAI), a framework that enables users to manually compress an existing conversation through two independent but unified entry points: pre-open compression, initiated directly from the conversation history interface without opening or rendering the complete original conversation, and in-conversation compression, initiated after the conversation has been opened. Both entry points operate under the same user-defined compression configuration rather than relying on automatic device-adaptive decisions. Users can determine the compression scope, intensity, transformation method, preservation requirements, attachment treatment, and lifecycle of the original conversation. The framework distinguishes lossless technical compression from semantic and hybrid compression, maintains conversation identity and structural relationships, and introduces verification, version management, and recoverability mechanisms. Its central objective is to make large conversational objects more manageable without requiring users to abandon valuable historical interactions or automatically divide them into separate conversations. To the best of our knowledge, this paper presents one of the first explicitly integrated formulations of a user-controlled, dual-entry framework for persistent application-level conversation compression that treats pre-open and in-conversation compression as two operational pathways within one unified lifecycle. This is a scoped novelty claim rather than a claim to have invented conversation compression, dialogue summarization, or context reduction. The paper establishes the framework's conceptual foundations, formal model, system architecture, compression lifecycle, integrity requirements, and empirical evaluation agenda.
No takes yet. Share an insight, caveat, or question.
Jincheng Zhang (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: