Get Started
Home
Topics
Search
Library
Research questionHow can memory-augmented LLM agents coordinate memory construction, retrieval, and repair over long-horizon interactions?External memory can help LLM agents retain information across long interactions, but its construction and retrieval often rely on local heuristics. Failures in later agent responses are rarely converted into targeted repairs, leaving the memory bank unreliable.
AI
AI Agents
AI Memory
Information Retrieval
Machine Learning
Multi-agent Systems
Natural Language Processing
Reasoning
Research Paper
Retrieval-Augmented Generation
Latest papersRecent research connected to this question, newest first.MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-EvolutionApplies to memory-augmented LLM agents using external memory banks during long-horizon interaction. The evidence covers coordinated construction, iterative retrieval, and in-situ verification and repair, with experiments on LoCoMo across multiple LLM backbones and three storage backends.research paper · Sep 3, 2026Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM SystemsThe source models orchestrator-worker interaction as a bilevel coordination game and reflection as stochastic movement over semantic memory states. It distinguishes transcript-only gates from environment-grounded evaluation and gives convergence, rate, and re-anchoring results under stated assumptions. Empirical evidence uses environment-grounded metrics in a Kimi-based system on 500 SWE-bench instances, so conclusions beyond those assumptions and that task setting are not established.research paper · Sep 2, 2026AdaMem: Learning What to Remember with Adaptive Memory Policies for Personalized AgentsThe evidence concerns personalized LLM agents receiving periodic feedback about memory preferences. Evaluation uses six concurrent interaction personas across five ten-week stories, two extraction models, and two feedback modes; it measures question-answering accuracy and persistent-memory size, while also reporting difficulty executing learned policies reliably.research paper · Sep 2, 2026CHIME: Credit-Aware Hierarchical Memory Evolution for Long-Horizon Agentic PlanningApplies to agents that accumulate interaction experience in external memory while improving at inference time. The evidence covers four long-horizon agent benchmarks and does not establish performance in other settings.research paper · Sep 2, 2026
Related questions
How can LLM agents extract useful long-term memories for unknown future tasks without storing hallucinated facts?How can personalized LLM agents retrieve time-valid memories of persistent and evolving user states?How can LLM orchestrators preserve continuous state when collaborating with non-language agents?How can LLM agents reuse procedural skills across diverse long-horizon tasks without generic or instance-bound memory?