ReasoningBank: an agent memory framework that lets LLM agents keep learning from wins and losses
Summary
ReasoningBank gives LLM agents a memory layer so they continuously learn from both successes and failures, driving higher task success rates and faster execution.
View Cached Full Text
Cached at: 04/22/26, 03:00 PM
ReasoningBank is a novel agent-memory framework that enables large-language-model (LLM) agents to keep learning from both successes and failures. Our evaluation shows the framework boosts agent effectiveness, markedly raising task-success rates and work efficiency.
Similar Articles
@GoogleResearch: ReasoningBank, a novel agent memory framework, enables LLM agents to continuously learn from both successful & failed e…
Google Research introduces ReasoningBank, an agent memory framework that lets LLM agents learn continuously from successes and failures, improving success rates and efficiency.
@chenchengpro: The more fancy "memory" architectures you stack on an LLM Agent, the better the results? Not necessarily. A new paper tested 12 memory systems and found no universal winner. It decomposes Agent memory like a database — representation & storage, extraction, retrieval & routing, and maintenance — and tested Mem0, Letta, Zep, C…
A paper systematically evaluates 12 LLM Agent memory systems, breaks them into four modules, finds no single architecture dominates all scenarios, and reveals cost-performance trade-offs and common issues (e.g., 'past hallucinations').
@rohanpaul_ai: New Illinois+ Tsinghua University and other labs study finds that LLM agents still have unreliable memory and that it c…
A study by University of Illinois and Tsinghua University finds that LLM agents' memory becomes unreliable when continuously rewritten, with performance dropping from 100% to 54% on ARC-AGI tasks. The paper proposes preserving raw episodes instead of always summarizing them.
PolicyBank: Evolving Policy Understanding for LLM Agents
PolicyBank proposes a memory mechanism that enables LLM agents to autonomously refine their understanding of organizational policies through iterative interaction and corrective feedback, closing specification gaps that cause systematic behavioral divergence from true requirements. The work introduces a systematic testbed and demonstrates PolicyBank can close up to 82% of policy-gap alignment failures, significantly outperforming existing memory mechanisms.
Human-Inspired Memory Architecture for LLM Agents
Microsoft researchers propose a biologically-inspired memory architecture for LLM agents that incorporates mechanisms like sleep-phase consolidation and interference-based forgetting to manage persistent memory efficiently.