ACE: Pluggable Adaptive Context Elasticizer across Agents

2026-06-30

AI Summary

This work addresses the challenge of excessively long reasoning trajectories in large language model (LLM) agents, which are constrained by fixed context windows and suffer from irreversible information loss when using conventional truncation or summarization methods. To overcome this limitation, the authors propose ACE (Adaptive Context Elasticity), a plug-and-play module featuring a reversible, lossless message preservation layer and a task-state-aware dynamic orchestration mechanism. ACE adaptively selects historical steps at each decision point, incorporating them into the context in their original form, as summaries, or omitting them entirely—without requiring any training. The module seamlessly integrates with diverse agent frameworks such as ReAct, DeepAgent, WebThinker, and MiroFlow, consistently outperforming existing baselines across all four while preserving informational integrity and enhancing context efficiency.

Abstract

The increasing complexity of agentic tasks has led to rapidly growing trajectory lengths, which poses significant challenges for large language model (LLM) based agents with fixed context windows. Existing context management techniques, such as truncation and summarization, suffer from inherent inflexibility and irreversibility: once information is discarded or compressed, it cannot be recovered even when it becomes critically relevant in later decision steps. To address these limitations, we propose the Adaptive Context Elasticizer (ACE), a plug-and-play module that elastically orchestrates historical step information into the agent's context at each decision step. ACE maintains a lossless message maintenance layer that stores both raw messages and compressed abstractions for each historical step, while a context orchestration layer adaptively assigns each step an elastic type as raw, abstract, or drop, at every decision step based on the current task state. This reversible design ensures that the main LLM always receives a compact yet information-rich context. We adapt ACE to four diverse agent frameworks, including ReAct, DeepAgent, WebThinker, and MiroFlow, without training or architectural modifications. Experiments show that ACE consistently outperforms truncation and summarization baselines, and brings consistent performance gains across all four agent frameworks.

Problem

• context management
• large language model agents
• trajectory length
• irreversibility
• fixed context window

Innovation

• Adaptive Context Elasticizer
• lossless context management
• elastic context orchestration
• plug-and-play agent module
• reversible summarization