Multimodal Agent MemorySubmit a paper

AgentOCR

AgentOCR: Optical Context Compression for Long-Horizon Multimodal Agents

Lang Feng, Fuchao Yang, Feng Chen, Xin Cheng, Haiyang Xu, Zhenglin Wan, et al. · 2026-01-08

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. optically cached rendered interaction history

    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
    Accumulated textual/action history is rendered into compact image segments and optically cached.Full paper: Optical Memory Encoding · PDF p. 4 · Figure 1

How memory changes (T2 operations)

  • Update · Rule-based update

    During operation, AgentOCR applies fixed-rule insertion or replacement to optically cached rendered interaction history; the memory state changes while the governing update policy remains fixed.

  • Update · Learned update module

    During operation, AgentOCR applies a trained but frozen update module to optically cached rendered interaction history; the memory state changes while the governing update policy remains fixed.

Related systems