Multimodal Agent MemorySubmit a paper

VimRAG

VimRAG: Navigating Massive Visual Context in Retrieval-Augmented Generation via Multimodal Memory Graph

Qiuchen Wang, Shihang Wang, Yu Zeng, Qiang Zhang, Fanrui Zhang, Zhuoning Guo, et al. · 2026-04-21

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Image
Also: Video & streaming, text
Representation
Hybrid
graph-linked visual evidence and structured reasoning state

How experience becomes memory

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

  1. multimodal evidence nodes

    1. Represented asRetained perceptual source evidenceSource- and modality-specific
    2. Stored inGraph / relational store, Vector-indexed collection
    3. Retrieved byStructure traversal, Planner / tool-assisted retrieval
    4. Source evidenceSource directly available
    The directed acyclic memory graph preserves retrieved image/video evidence.Full paper: Policy Model · PDF p. 2

  2. structured agent states and evidence graph

    1. Represented asSpatial, temporal & cross-modal structured memoryCross-represented explicit
    2. Stored inGraph / relational store
    3. Retrieved byStructure traversal
    4. Source evidenceSource linked & recoverable
    Reasoning states and retrieved multimodal evidence are organized in a dynamic DAG.Full-paper method / memory representation and retrieval · PDF p. 5

How memory changes (T2 operations)

  • Update · Prompted LLM/VLM record update

    During operation, VimRAG applies LLM-mediated memory editing to multimodal evidence nodes; the memory state changes while the governing update policy remains fixed.

Related systems