arXiv cs.AI
·
18 hours ago
ECHO is a selective turn-memory framework for language agents in reinforcement learning that compresses environment interactions into indexed records and reconstructs context windows while preserving traceability for credit assignment. On BrowseComp-Plus, ECHO achieved 43.4% accuracy compared to 28.9% for GRPO and 36.1% for the rolling-summary baseline SUPO. The approach enables policies to improve credit assignment to supporting evidence while using fewer turns and lower trajectory volume than competing methods.
arXiv cs.AI
·
18 hours ago
Researchers introduced MemCon, a framework that models memory operations in LLM agents as a Markov Decision Process to learn adaptive policies for when and how to retrieve, inject, or consolidate memories. The method achieved up to 15.2 point improvements in task success across 6 benchmarks while reducing token consumption by 5-20 percent. This allows memory management to adjust dynamically based on task context rather than relying on fixed heuristics.
arXiv cs.AI
·
18 hours ago
Researchers developed HealthClaw, an open-source AI agent architecture that maintains longitudinal memory of personal health information and updates its support as a person's health status changes over time. Across 900 longitudinal support probes, the agent achieved 45.7% answer accuracy compared to 0.2% with single-query prompting, while using 71.7% less context than full-history methods. The system balances personalized health management with privacy protection, though the researchers note that clinical effectiveness requires real-world prospective evaluation.
arXiv cs.AI
·
18 hours ago
Oracle developed Agent Memory, a database-native memory system built on Oracle Database designed to manage long-horizon AI agents by retaining task state across conversations and accumulating procedural knowledge. The system achieved 93.8% accuracy on LongMemEval benchmarks while using approximately 10.7 times fewer tokens than flat-history baselines through a layered architecture separating active and passive memory. The implementation enables practical enterprise deployments of AI agents by providing scoped memory retrieval with explicit controls across users, agents, and threads.
