Multimodal Agent MemorySubmit a paper

MemOCR

MemOCR: Layout-Aware Visual Memory for Efficient Long-Horizon Reasoning

Yaorui Shi, Shugui Liu, Yu Yang, Wenyu Mao, Yuxin Chen, Qi Gu, et al. · 2026-01-29

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Image
Also: text
Representation
Cross-represented explicit

How experience becomes memory

Each path is one memory component, traced from what is stored to how it is found again.

  1. layout-aware rich-text visual memory

    1. Represented asPerception-to-text records & summariesCross-represented explicit
    2. Stored inHierarchical multi-tier store
    3. Retrieved byPlanner / tool-assisted retrieval
    4. Source evidenceSource not retained
    The system maintains structured rich-text history and renders an adaptively compressed visual memory for later reasoning.Full paper: 3. Method: MemOCR · PDF p. 3 · Figure 1

How memory changes (T2 operations)

  • Update · Prompted LLM/VLM record update

    The memory writer rewrites accumulated content under a fixed budget-aware compression policy. For MemOCR, this specifically supports LLM-mediated memory editing of layout-aware rich-text visual memory.

  • Consolidation · Episodic / hierarchical summarization

    Older interaction content is compressed into denser layout-aware memory. For MemOCR, this specifically supports episodic or hierarchical summarization of layout-aware rich-text visual memory.

Related systems