Show HN: HMLR – AI Memory system that gets 1.00/1.00 on every impossible test (github.com)

HMLR, or Hierarchical Memory Lookup & Routing, introduces a groundbreaking long-term memory architecture for AI agents, achieving unprecedented scores of 1.00 in both faithfulness and context recall across multiple complex benchmark tests. Unlike conventional systems that rely on large context windows and simplistic retrieval methods, HMLR employs a structured, state-aware memory framework that can manage conflicting information over time while ensuring policy compliance and multi-hop reasoning, all using mini-tier models like GPT-4.1-mini. This development is significant for the AI/ML community as it addresses long-standing issues in memory systems, such as temporal truth resolution, user invariance across topics, and the ability to retrieve relevant information without keyword overlap. The architecture's performance not only showcases the effectiveness of a well-designed memory system over sheer model size but also opens the door for more efficient AI entities capable of maintaining persistent memory and governance. With its open-source nature and reproducible benchmarks, HMLR stands to inspire future innovations in AI memory architectures and could reshape how agents interact with users over extended periods.