Show HN: I built an agent memory system for myself and got 90.8% on LongMemEval

Category: ai-ml

Tags: ai-agents, knowledge-graph, memory-system

Score: 7.3/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 7)

Memento is a bitemporal knowledge graph system for AI agents that provides structured, persistent memory by tracking when facts were true versus when they were learned. It's interesting because it replaces simple vector store retrieval with entity resolution, contradiction detection, and graph-based reasoning, achieving 90.8% accuracy on the LongMemEval benchmark for long-term conversational memory.

Target audience: ai-engineers, backend-devs

Repository: https://github.com/shane-farkas/memento-memory · Python · MIT

View on Hacker News