Methods reduce communication in multiagent AI systems for faster results
Selective and Hybrid-State Communication for Latent Multi-Agent Systems
Multiagent SystemsMachine Learning
Summary
Communicating between AI agents can be slow and costly when sharing lots of details repeatedly. The authors present two techniques to cut down this communication by sharing only necessary internal information or combining different model states efficiently. One method, SelectKV, carefully chooses what internal data is shared to save bandwidth without losing accuracy. The other, HyLaMAS, works on newer AI models by sharing mixed types of information and speeds up the process significantly. Together, these methods help AI systems that work together do so more quickly and efficiently.
What this means in practice
- •For ai system developers: Reduce communication overhead between cooperating AI agents to speed up distributed model inference using latent state exchange.
- •For cloud service engineers: Cut token processing and network load in multi-agent AI services for faster user responses by transmitting compressed internal model states.
Authors
Miles Liao, Aaron Smieszek, Yuhao Ge
Abstract
Efficient communication remains a major bottleneck in large language model multiagent systems. Latent communication reduces repeated natural-language exchange by transferring internal model states between agents, but these transfers remain costly and are typically designed around conventional KV-cache representations. We address this problem from two complementary perspectives. We introduce SelectKV, a boundary-aware method that selectively communicates KV states while protecting inherited context, and HyLaMAS, which extends latent collaboration to hybrid linear-attention models by communicating both KV caches and recurrent states through a shared workspace. Across six benchmarks, SelectKV reduces communicated KV positions and logical KV payload by 9.7-9.9%, with accuracy unchanged on two benchmarks, within one percentage point on three, and six points lower on HumanEval+. Across Qwen3.5 model scales, HyLaMAS reduces decoded output tokens by 53-80% and achieves 1.16-2.78x speedups relative to text-based multi-agent communication. Together, these independently evaluated methods represent two complementary approaches to efficient latent-state communication across conventional Transformer and hybrid model architectures.