Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

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

Source

arXiv

arxiv.org/abs/2606.31650v1

Read the source

Source

Plain-Language Brief

The paper proposes ECHO, a selective turn-memory framework for long-horizon language agents. ECHO aims to address the issues of history collapse and traceable learning by compressing completed environment turns into compact memory records, reconstructing policy contexts from these records, and using source indices to route positive outcome credit to supporting evidence and actions. The authors report that ECHO outperforms existing methods on a multi-task benchmark and uses fewer turns and less trajectory volume.

What the Document Says

  • The paper proposes a framework called ECHO for long-horizon language agents.
  • ECHO addresses history collapse and traceable learning issues in existing context-management methods.
  • ECHO compresses completed environment turns into compact memory records.
  • ECHO reconstructs policy contexts by selecting from these memory records.
  • ECHO routes positive outcome credit to the evidence and selection actions that support successful answers using source indices.
  • On BrowseComp-Plus, ECHO outperforms GRPO and a rolling-summary baseline SUPO with fewer turns and lower trajectory volume.

Key Concepts Mentioned

  • long-horizon language agents
  • context-management methods
  • history collapse
  • traceable learning
  • selective turn-memory framework
  • ECHO
  • BrowseComp-Plus
  • GRPO
  • SUPO

Claims Made by the Document

  • ECHO outperforms existing methods on a multi-task benchmark.
  • ECHO uses fewer turns and less trajectory volume than SUPO.

Evidence Used

  • 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)

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.