Multimodal Agent MemorySubmit a paper

StreamFlow

StreamFlow: Dynamic Memory Flows for Streaming Video Understanding

Muxin Fu, Yifan Zhang, Wentao Zhang, Fangming Guo, Qian Chen, Guibin Zhang, et al. · 2026-08-11

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Video & streaming
Also: text
Representation
Hybrid
Distinct operative representation paths are explicitly linked for downstream retrieval, reasoning, or control.

How experience becomes memory

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

  1. dynamics-aware mid-term visual memory

    1. Represented asModality-specific encodings & geometric stateSource- and modality-specific
    2. Stored inHierarchical multi-tier store
    3. Retrieved byMetadata / spatiotemporal retrieval
    4. Source evidenceRetained but not accessible to the agent
    A lightweight mid-term memory filters temporal redundancy before full visual encoding.Full paper: 3.3. Latent Long-Term Memory · PDF p. 4 · Figure 1

  2. latent long-term visual memory

    1. Represented asPersistent multimodal latent stateInternalized latent & parametric
    2. Stored inHierarchical multi-tier store
    3. Retrieved bySimilarity search, Latent / parametric activation
    4. Source evidenceRetained but not accessible to the agent
    Historical content is consolidated into visual latents and injected by attention-guided retrieval.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1

How memory changes (T2 operations)

  • Update · Rule-based update

    During operation, StreamFlow applies fixed-rule insertion or replacement to dynamics-aware mid-term visual memory; the memory state changes while the governing update policy remains fixed.

  • Consolidation · Merging & deduplication

    Redundant streaming observations are filtered before encoding. For StreamFlow, this specifically supports merging or deduplication of overlapping entries of dynamics-aware mid-term visual memory.

  • Consolidation · Episodic / hierarchical summarization

    Mid-term content is consolidated into long-term visual latents. For StreamFlow, this specifically supports episodic or hierarchical summarization of dynamics-aware mid-term visual memory.

Related systems