Reproduce “KITE: Retrieval as a Decision Procedure for Long-Term Conversational Memory” 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

Feng Li, Kailun Tan, Zhibo Zhou, Youxiang Wang, Lejun Liu, Huaigu Li, Fei Du, Huilong An

KITE addresses long-term conversational retrieval by organizing source-linked facts into a shared index and compiling questions into validated conditions, rather than relying only on semantic similarity. On LongMemEval-S, it reaches 86.0% accuracy with 1.6k reader-context tokens, versus EverMemOS’s 83.0% and 2.8k. On fixed-memory LoCoMo, it also outperforms hybrid retrieval; correctness still depends on extracted memories and plans.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.