Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice

查看原文 / View Original 编辑 / Edit

Computer Science > Artificial Intelligence arXiv:2610.10590 (cs) [Submitted on 6 Oct 2026] Title: Agent-Controlled Forgetting for Tool-Using Agents: Reversible Context Curation in Practice Authors: Jan-Peter Franke.

计算机科学 > 人工智能 arXiv:2610.10590 (cs) [提交于 2026 年 10 月 6 日] 标题：面向工具使用智能体的智能体可控遗忘：可逆上下文管理的实践作者：Jan-Peter Franke。

Abstract: Tool-using agents repeatedly carry observations whose useful content can be much smaller than their original payload. We study agent-controlled forgetting: the acting model selects previously observed tool results, replaces each with a short note at its original position, and retains the exact original in a recoverable archive.

摘要：使用工具的智能体会反复携带观测结果，而这些观测结果的有用内容可能远小于其原始负载。我们研究了智能体可控遗忘：执行模型选择先前观测到的工具结果，在原始位置用简短注释替换每一个结果，并将确切的原始数据保留在可恢复的存档中。

A Python harness exposes batch archival and explicit recovery without task-specific model training, while protecting user instructions and assistant messages from these operations. In an exploratory OpenTelemetry debugging case followed by an unrelated implementation task, the method ended with 231,951 provider-reported prompt tokens versus 912,492 under retained history, used 50% fewer cumulative input tokens, and had an estimated API cost of USD 1.28-1.44 versus approximately USD 4.38.

一个 Python 工具包在无需特定任务模型训练的情况下，提供了批量归档和显式恢复功能，同时保护用户指令和助手消息免受这些操作的影响。在一个探索性的 OpenTelemetry 调试案例及其后的无关实现任务中，该方法最终产生的提供商报告提示词元（tokens）为 231,951 个，而保留历史记录的情况下为 912,492 个，累计输入词元减少了 50%，估计 API 成本为 1.28-1.44 美元，而对比组约为 4.38 美元。

Both arms passed the two-case primary behavioral oracle; neither fully satisfied the follow-up evaluation. The method made more requests and took 17% longer. A contrasting application-development pair produced no context or cost saving, and an earlier continuation exhibited lower manually assessed quality despite reduced context. These observations demonstrate substantial resource savings in noisy tool-use trajectories and identify workload dependence as a central consideration for reversible context management.

两组实验均通过了双案例主要行为预言测试；但均未完全满足后续评估。该方法发出了更多请求，耗时增加了 17%。另一组对比的应用开发实验未产生上下文或成本节省，且早期的一次延续实验在上下文减少的情况下，人工评估质量反而较低。这些观察结果表明，在嘈杂的工具使用轨迹中，该方法能显著节省资源，并指出工作负载依赖性是可逆上下文管理的核心考量因素。