Universal Multimodal Multi-Conversation Continuation: Toward the First User-Directed Framework for Cross-Conversation and Cross-Modal Context Inheritance in Generative AI
Generative artificial intelligence (AI) systems increasingly support extended conversations involving text, images, audio, video, and external documents. However, conversational interaction is still commonly organized around individual conversation threads, creating a structural limitation when users wish to continue work distributed across multiple previously established conversations. Existing approaches to conversation branching, conversational memory, multi-session dialogue, multimodal interaction, and conversation referencing address important aspects of this problem, but they do not necessarily provide a unified mechanism through which users can manually select multiple independent historical conversations, initiate an entirely new conversation, and inherit their multimodal content and associated contextual structures as a coherent foundation for subsequent interaction. This paper introduces Universal Multimodal Multi-Conversation Continuation (UMMCC), a proposed framework that addresses this problem through user-directed, multi-source, multimodal conversational context inheritance. The central principle is that historical conversations should not remain isolated interaction records; instead, they should become selectively reusable contextual sources from which new conversational environments can be created without modifying their original threads. UMMCC integrates seven conceptual capabilities: user-controlled conversation selection, multi-source context inheritance, multimodal asset inheritance, temporal and structural reconstruction, cross-conversation relationship preservation, provenance-aware context management, and independent new-conversation initialization. The framework is designed to accommodate textual messages, images, video, audio, PDF files, office documents, and other supported data types, subject to the underlying model's modality-processing capabilities. Unlike conventional conversation branching, which typically derives a new thread from a particular point in one existing conversation, UMMCC considers multiple independently selected conversations as potential parent sources of a new conversational context. Unlike simple conversation concatenation, it emphasizes the preservation of source identity, message relationships, temporal information, and cross-modal associations. This paper presents the conceptual foundations, system architecture, operational workflow, design requirements, potential applications, and evaluation methodology of UMMCC. To the best of our knowledge, this paper proposes one of the first explicitly integrated frameworks centered on user-directed continuation from multiple historical conversations with multimodal context inheritance and provenance preservation. This is a scoped priority claim rather than a claim of exhaustive proof of worldwide novelty. No experimental superiority or implementation results are claimed.
Authors
- Jincheng Zhang (ORCID: https://orcid.org/0000-0001-7332-6704)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-13
- DOI
- https://doi.org/10.5281/zenodo.24184470
- Primary Topic
- Speech and dialogue systems
- Type
- preprint