Multimodal Agent MemorySubmit a paper

MuKV

MuKV: Multi-Granular Key-Value Cache for Efficient Long Video Understanding

Junbin Xiao, Jiajun Chen, Tianxiang Sun, Xun Yang, Angela Yao · 2026-05-21

Read the paperPDF

Management
T2 Dynamic Memory
Memory changes during deployment through a stable management policy.
Primary modality
Video & streaming
Also: text
Representation
Internalized latent & parametric

How experience becomes memory

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

  1. offline multigranular visual KV cache

    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
    An offline multi-grained KV cache is queried through semi-hierarchical retrieval.Full-paper method / memory representation and retrieval · PDF p. 2 · Figure 1

How memory changes (T2 operations)

  • Update · Rule-based update

    During operation, MuKV applies fixed-rule insertion or replacement to offline multigranular visual KV cache; the memory state changes while the governing update policy remains fixed.

  • Consolidation · Merging & deduplication

    During operation, MuKV applies merging or deduplication of overlapping entries to offline multigranular visual KV cache; the memory state changes while the governing update policy remains fixed.

  • Forgetting · Discrete eviction

    During operation, MuKV applies explicit deletion or eviction of stored items to offline multigranular visual KV cache; the memory state changes while the governing update policy remains fixed.

Related systems