Reproduce “Zep: A Temporal Knowledge Graph Architecture for 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
Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan, Daniel Chalef
ZEP presents a temporal knowledge graph architecture designed to enhance AI agents with an external memory system that maintains temporal awareness and structured knowledge. This approach helps overcome LLM context window limitations by achieving 94.8% accuracy on the Deep Memory Retrieval benchmark and demonstrating up to 18.5% accuracy improvement with up to 90% latency reduction on the LongMemEval benchmark.
Key insights
- A temporal knowledge graph architecture can effectively address the limitations of LLM context windows by enabling robust, time-aware memory for AI agents.
- Organizing knowledge into hierarchical subgraphs (episode, entity, community) allows for efficient organization and retrieval across different levels of abstraction.
- Integrating multi-faceted search (semantic, full-text, graph traversal) combined with re-ranking significantly improves information retrieval from dynamic knowledge graphs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.