ECHO: Prune to act, trace to learn with selective turn memory in agentic RL

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Zijun Xie, Binbin Zheng, Enlei Gong, Jihua Liu, Yuyang You, Lingfeng Liu, Jiayao Tang, Guanqun Zhao, Aoqi Hu, Zeyu Chen
• Venue: arXiv: cs.LG
• Published: 2026-06-30
• Collected: 2026-07-01
• URL: http://arxiv.org/abs/2606.31650v1
• PDF: https://arxiv.org/pdf/2606.31650v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers propose ECHO, a selective turn-memory framework to address limitations of existing context management methods in long-horizon language agents. ECHO compresses completed environment turns into compact memory records, reconstructs bounded policy contexts by selecting from these records, and routes positive outcome credit back to the evidence and actions that support successful answers. On BrowseComp-Plus, ECHO outperforms other methods, using fewer turns and lower trajectory volume.

What the Document Says

• The paper introduces ECHO, a selective turn-memory framework for long-horizon language agents.
• ECHO compresses completed environment turns into compact memory records.
• ECHO reconstructs bounded policy contexts by selecting from these records.
• ECHO routes positive outcome credit to the evidence and selection actions that support successful answers.
• On BrowseComp-Plus, ECHO outperforms GRPO (28.9%) and the rolling-summary baseline SUPO (36.1%).
• ECHO uses fewer turns and lower trajectory volume than SUPO.

Key Concepts Mentioned

• long-horizon language agents
• context management methods
• selective turn-memory framework
• ECHO
• compact memory records
• bounded policy contexts
• positive outcome credit
• BrowseComp-Plus

Claims Made by the Document

• ECHO addresses limitations of existing context management methods.
• ECHO outperforms GRPO and the rolling-summary baseline SUPO on BrowseComp-Plus.

Evidence Used

• The paper describes ECHO as a solution to coupled limitations of existing methods.
• Results show that ECHO outperforms other methods on BrowseComp-Plus.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Long-horizon language agents must repeatedly interact with tools, accumulate evidence, and make decisions under bounded context windows. Existing context-management methods make such rollouts feasible by truncating distant history, folding past turns into summaries, or selecting compact memory states. However, these breakthroughs introduce two coupled limitations. First, as the number of turns grows, historical observations are progressively removed or collapsed into compressed states, making it harder for the policy to reuse fine-grained evidence. Second, once the original turns are no longer source-addressable, outcome-based RL loses an explicit path for aligning policy updates with the evidence that supported a successful final answer. To this end, we propose ECHO, a selective turn-memory framework that jointly addresses history collapse and traceable learning through source-indexed reconstruction. Specifically, ECHO compresses each completed environment turn into a compact memory record, reconstructs bounded policy contexts by selecting from these records, and reuses the selected source indices to route positive outcome credit to the evidence and selection actions that support successful answers. On BrowseComp-Plus, ECHO reaches 43.4% held-out accuracy, outperforming GRPO (28.9%) and the rolling-summary baseline SUPO (36.1%), while using fewer turns and lower trajectory volume than SUPO (Figure 1). Additionally, the trained policy improves zero-shot generalization across multi-objective QA, code generation, and deep information-seeking benchmarks on both dense and MoE backbones.