Efficient long-horizon agents prune history safely to save compute
DRSR: Learning Set-Level Deletion Risk for Efficient Long-Horizon Agents
Artificial Intelligence
Summary
When AI agents handle tasks over long periods, they collect lots of past information, but not all of it stays important. The authors found that just scoring pieces of history alone isn't enough to safely throw some of it away. They created a new method, DRSR, which looks at groups of past information together to decide what can be removed without harming the agent's performance. This approach helps the AI save processing power while still making good decisions.
What this means in practice
- •For ai system engineers: Reduce the token usage of language-model agents by safely pruning historical context without harming performance.
- •For cloud service operators: Lower computational costs by implementing set-level risk pruning to compress agent histories for long-running AI tasks.
Authors
Mingxuan Wang, Bo Wang, Fei Luo, Guorun Yao, Chao Ning, Yinglong Guo, Hongyue Chen, Yanbiao Ma, Jungong Han
Abstract
Long-horizon language-model agents accumulate reasoning traces, tool exchanges, and observations whose relevance changes with the current decision. Existing compression strategies often score historical units independently, but the safety of deleting several units is generally not determined by their singleton scores: redundant evidence, accumulated small effects, and the information that remains after deletion all matter. We introduce Direct Relational Set-Risk Pruning (DRSR), which formulates agent-history compression as risk-constrained selection over deletion sets. Offline, DRSR constructs exact counterfactual supervision by jointly deleting protocol-valid history Blocks and measuring the change in teacher-forced likelihood of the same recorded next output. A lightweight scorer then predicts set-level harm from online-visible relations between candidate history and the current pre-action state, together with deleted-retained and pairwise set structure. At deployment, DRSR evaluates a small set of structurally valid deletion candidates with the lightweight scorer and removes the largest feasible set under recency, protocol, budget, and learned-risk constraints, abstaining when no set is sufficiently safe. On WorkBuddyBench Full260, DRSR increases mean reward from 0.699 to 0.802 while reducing total model tokens by 20.820%. On the fixed Eval40 comparison, it obtains 0.794 reward at 1.211M tokens per task, using 35.850% fewer tokens than the uncompressed agent. Mechanistic analyses and ablations further show that decision-conditioned relations, retained-context information, pair interactions, and abstention each contribute to reliable pruning.