Reproduce “MEMOIR: A Training-Free Memory Lifecycle for Long-Horizon LLM Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mayank Anand, Subham Raj, Aman Vaibhav Jha, Sriparna Saha
MEMOIR links retrieval feedback with screened memory compression for assistants handling long conversations. After each answer, it estimates which retrieved notes were useful and updates their importance; proposed merges must pass preservation and retrieval-coverage checks. On sequential queries across ten shared LoCoMo stores, confidence-gated TRACE-UCB raised evidence recall@5 by 1.50 points, not answer accuracy or fresh-store performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.