mlsanityData to 14 Oct 2026

Stable Geometry with Divergent Task Evidence for Efficient Long-Horizon Agent Compression

arXiv 2609.27332 ↗First posted 23 Sep 2026cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Long horizon agents accumulate growing interaction histories that increase context and inference costs. We find that geometric redundancy alone is an insufficient criterion for safe compression. Although agent histories exhibit strong low dimensional structure, similar global geometry can preserve very different amounts of task evidence. At identical retained block counts, evidence aware selection raises next action Top 3 retention from 0.31 to 0.69, while centroid similarity remains 0.98. Controlled replacement further shows that action related information can be substantially altered while global geometric measures remain nearly unchanged. Motivated by this gap between geometry and evidence, we introduce Geometry Guided Evidence Preserving Memory (GEM), a training free compressor that protects task and execution evidence before using geometric residuals to complete coverage. GEM reduces mean combined token usage from 2.69M to 2.11M per task, a 21.4% reduction, while maintaining comparable task reward. Our results show that efficient agent history compression should optimize for preserved task evidence rather than geometric coverage alone.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 36 of its 39 references that are in the map. First posted on arXiv 23 Sep 2026.

Builds on

36 of 39 references.

References · 36

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  3. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  4. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,568
  5. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  6. 2021SimCSE: Simple Contrastive Learning of Sentence Embeddingscited by 2,173
  7. 2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflectioncited by 1,839
  8. 2023LongBench: A Bilingual, Multitask Benchmark for Long Context Understandingcited by 1,417
  9. 2023MemGPT: Towards LLMs as Operating Systemscited by 1,389
  10. 2024RULER: What's the Real Context Size of Your Long-Context Language Models?cited by 1,346
  11. 2023AgentBench: Evaluating LLMs as Agentscited by 1,267
  12. 2026DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligencecited by 1,040
24 more
  1. 2019How Contextual are Contextualized Word Representations? Comparing the Geometry of BERT, ELMo, and GPT-2 Embeddingscited by 784
  2. 2024RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrievalcited by 532
  3. 2023LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compressioncited by 453
  4. 2024Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexitycited by 451
  5. 2023Learning to Compress Prompts with Gist Tokenscited by 390
  6. 2024LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitaskscited by 379
  7. 2024∞Bench: Extending Long Context Evaluation Beyond 100K Tokenscited by 367
  8. 2024LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compressioncited by 348
  9. 2025MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agentscited by 238
  10. 2017All-but-the-Top: Simple and Effective Postprocessing for Word Representationscited by 212
  11. 2023In-context Autoencoder for Context Compression in a Large Language Modelcited by 199
  12. 2025ACON: Optimizing Context Compression for Long-horizon LLM Agentscited by 136
  13. 2024LazyLLM: Dynamic Token Pruning for Efficient Long Context LLM Inferencecited by 70
  14. 2026SWE-Pruner: Self-Adaptive Context Pruning for Coding Agentscited by 44
  15. 2025Sculptor: Empowering LLMs with Cognitive Agency via Active Context Managementcited by 18
  16. 2026Beyond Semantic Organization: Memory as Execution State Management for Long-Horizon Agentscited by 11
  17. 2026Self-Compacting Language Model Agentscited by 11
  18. 2026Self-GC: Self-Governing Context for Long-Horizon LLM Agentscited by 11
  19. 2026Remember When It Matters: Proactive Memory Agent for Long-Horizon Agentscited by 11
  20. 2026SAM: State-Adaptive Memory for Long-Horizon Reasoning Agentcited by 10
  21. 2026ACM: Agentic Context Management for Long Horizon Taskscited by 10
  22. 2026Tencent WorkBuddy Bench: A Multi-Domain Coding-Agent Benchmark with Contamination-Resistant Task Constructioncited by 6
  23. 2026CoMem: Context Management with A Decoupled Long-Context Modelcited by 5
  24. 2026Learning What Not to Forget: Long-Horizon Agent Memory from a Few Kilobytes of Learningcited by 4

Built on by

Not cited yet.