AgentSwing: Adaptive Parallel Context Management Routing for Long-Horizon Web Agents

2026-03-28

AI Summary

This work addresses the challenge faced by long-horizon web agents operating under limited context windows, where static context management struggles to adapt to dynamic task requirements. To overcome this limitation, the authors propose a state-aware, adaptive parallel context management routing framework. At each decision point, the method concurrently expands multiple context branches and employs a lookahead routing mechanism to select the optimal continuation path. Notably, it introduces a probabilistic framework to model the success dimensions of long-horizon tasks, enabling dynamic, parallel, and adaptive context management. Experimental results demonstrate that the proposed approach significantly outperforms existing static strategies across diverse benchmarks and agent backbones, reducing interaction steps by up to threefold while simultaneously elevating the upper bound of task performance.

Abstract

As large language models (LLMs) evolve into autonomous agents for long-horizon information-seeking, managing finite context capacity has become a critical bottleneck. Existing context management methods typically commit to a single fixed strategy throughout the entire trajectory. Such static designs may work well in some states, but they cannot adapt as the usefulness and reliability of the accumulated context evolve during long-horizon search. To formalize this challenge, we introduce a probabilistic framework that characterizes long-horizon success through two complementary dimensions: search efficiency and terminal precision. Building on this perspective, we propose AgentSwing, a state-aware adaptive parallel context management routing framework. At each trigger point, AgentSwing expands multiple context-managed branches in parallel and uses lookahead routing to select the most promising continuation. Experiments across diverse benchmarks and agent backbones show that AgentSwing consistently outperforms strong static context management methods, often matching or exceeding their performance with up to $3\times$ fewer interaction turns while also improving the ultimate performance ceiling of long-horizon web agents. Beyond the empirical gains, the proposed probabilistic framework provides a principled lens for analyzing and designing future context management strategies for long-horizon agents.

Problem

• context management
• long-horizon agents
• adaptive routing
• information-seeking
• LLM agents

Innovation

• adaptive context management
• long-horizon agents
• parallel routing
• lookahead selection
• probabilistic framework