mlsanityData to 7 Oct 2026

Receiver-Conditioned Latent Communication gives 94% CacheBack

arXiv 2609.32046 ↗First posted 25 Sep 2026cs.AIcited by 0

On the map: Efficient LLM inference and serving → Efficient reasoning and agent inference

Abstract (arXiv)

Multi-agent systems distribute large contexts across agents that communicate to solve a task. Text messages are compact but require decoding and may omit evidence the receiving agent needs. Recent latent communication instead transfers KV caches. This avoids text generation and can improve accuracy and latency. However, a full KV cache grows linearly with both the context an individual agent processes, and the number of agents that coordinate together. This raises memory and context costs, often far exceeding available GPU resources and context window sizes. Our key observation is that agents need only send what the receiving agent requires for its local task -- which we call receiver-conditioned communication. The receiver agent passes the sender a small description of its information needs, which serves to filter and compress the sender agent's KV cache. CacheBack is a simple, robust, training-free instance of receiver conditioning based on the sender's attention weights. On FanOutQA, CacheBack with Qwen 3 removes 75% of the state the agent would otherwise receive, improving accuracy by 14.7 percentage points and reducing median task-completion latency by 3.2x relative to text communication. We show comparable improvements across model families that span dense Transformers, Mamba-attention hybrids, and sliding-window attention.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Efficient reasoning and agent inference, one of 12 directions in Efficient LLM inference and serving. Builds on 23 of its 27 references that are in the map. First posted on arXiv 25 Sep 2026.

Builds on

23 of 27 references.

Closest ancestors · 2

Same direction, newest first

  1. 2026Beyond tokens: a unified framework for latent communication in LLM-based multi-agent systems4 Jun
  2. 2025Q-KVComm: Efficient Multi-Agent Communication Via Adaptive KV Cache Compression27 Nov

Other references · 21

  1. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
  2. 2024SnapKV: LLM Knows What You are Looking for Before Generationcited by 848
  3. 2023H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Modelscited by 805
  4. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  5. 2024LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitaskscited by 367
  6. 2024Magentic-One: A Generalist Multi-Agent System for Solving Complex Taskscited by 221
  7. 2024Chain of Agents: Large Language Models Collaborating on Long-Context Taskscited by 201
  8. 2025KVzip: Query-Agnostic KV Cache Compression with Context Reconstructioncited by 89
  9. 2025Latent Collaboration in Multi-Agent Systemscited by 67
  10. 2025ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inferencecited by 57
  11. 2025Cache-to-Cache: Direct Semantic Communication Between Large Language Modelscited by 54
  12. 2023Let Models Speak Ciphers: Multiagent Debate through Embeddingscited by 49
9 more
  1. 2024Optima: Optimizing Effectiveness and Efficiency for LLM-Based Multi-Agent Systemcited by 40
  2. 2025Enabling Agents to Communicate Entirely in Latent Spacecited by 31
  3. 2024FanOutQA: A Multi-Hop, Multi-Document Question Answering Benchmark for Large Language Modelscited by 29
  4. 2025Thought Communication in Multiagent Collaborationcited by 28
  5. 2025KVComm: Enabling Efficient LLM Communication through Selective KV Sharingcited by 16
  6. 2025Communicating Activations Between Language Model Agentscited by 15
  7. 2026GraphPlanner: Graph Memory-Augmented Agentic Routing for Multi-Agent LLMscited by 10
  8. 2026Latent Cache Flow: Model-to-Model Communication Without Textcited by 6
  9. 2026When Less Latent Leads to Better Relay: Information-Preserving Compression for Latent Multi-Agent LLM Collaborationcited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026LCGuard: Latent Communication Guard for Safe KV Sharing in Multi-Agent Systems
  2. 2026Task-Oriented Key-Layer KV Communication for Efficient Latent Multi-Agent Collaboration
  3. 2025KVCOMM: Online Cross-context KV-cache Communication for Efficient LLM-based Multi-agent Systems
  4. 2026KVCMAS: Efficient KV cache Correction for Shared Context in Multi-Agent Systems
  5. 2026Agent Memory Below the Prompt: Persistent Q4 KV Cache for Multi-Agent LLM Inference on Edge Devices
  6. 2026Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs
  7. 2026Persistent Memory in Multi-Agent LLM Inference: What It Costs, What It Buys, and When You Can Tell
  8. 2026PReCache: Efficient KV Cache Sharing for Multi-LoRA Agents via Low-Rank Precomputation and Neutral Reconstruction