Reproduce “MemCode: Discrete Semantic Representations for Long-Term Agent 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
Xin Li, Liang Hu, Duoqian Miao, Ming Peng, Qi Zhang
MemCode helps conversational assistants retrieve earlier facts by rewriting dialogue into timestamped, self-contained memory units and indexing them with both vector embeddings and discrete codebook associations. On LoCoMo with GPT-4o-mini, a frozen 5,911-AMU store, and 20 candidates per retrieval path, it achieves 78.12% accuracy versus 76.75% with BM25, with gains concentrated in multi-hop and temporal questions, not all categories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.