Memory Control Signals Emerge Before Action in Long Horizon Agents
cs.AI
Submitted: 2026-09-23
Updated: 2026-09-23
Comments: 35 pages
License: http://creativecommons.org/licenses/by/4.0/
The gist: Long horizon language model agents continuously accumulate interaction history, increasing computational cost while making relevant information harder to preserve and reuse.
Terminology
Abstract
Long horizon language model agents continuously accumulate interaction history, increasing computational cost while making relevant information harder to preserve and reuse. Existing context management methods mainly focus on how to compress or retrieve history, but largely leave open whether the model itself already represents the need for these memory operations before they occur. We study the hidden state immediately before each agent action and find that compression and recall needs are already encoded in the model's internal representations. These signals cannot be explained by simple context length or interaction progress, and they exhibit distinct formation patterns across model depth. We further show that most memory decision information is preserved in a compact recent context, while selectively restored historical evidence complements the long range dependencies that recent context misses. Based on these findings, we propose Preaction Memory with Evidence Retrieval (PaMER), which combines state guided compression with external evidence retrieval. PaMER+ further introduces step level evidence selection to recover only the historical information required by the current task. Experiments on WorkBuddyBench, across multiple context management baselines and model backbones, show that our framework substantially reduces context consumption while maintaining competitive task performance.
Sources
- RULER: What's the Real Context Size of Your Long-Context Language Models?
- Self-Compacting Language Model Agents
- Active Context Compression: Autonomous Memory Management in LLM Agents
- SAM: State-Adaptive Memory for Long-Horizon Reasoning Agent
- CoMem: Context Management with A Decoupled Long-Context Model
- ACM: Agentic Context Management for Long Horizon Tasks
- ACON: Optimizing Context Compression for Long-horizon LLM Agents
- Tencent WorkBuddy Bench: A Multi-Domain Coding-Agent Benchmark with Contamination-Resistant Task Construction
- SWE-Pruner: Self-Adaptive Context Pruning for Coding Agents
- Self-GC: Self-Governing Context for Long-Horizon LLM Agents
- LLM Agents Are Latent Context Managers: Eliciting Self-Managed Context via State Proprioception
- MemGPT: Towards LLMs as Operating Systems
- Understanding intermediate layers using linear classifier probes
- Discovering Latent Knowledge in Language Models Without Supervision
- Language Models (Mostly) Know What They Know
- DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
- GLM-5: from Vibe Coding to Agentic Engineering
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems