Just shipped v0.1.2 of HMLR, an open-source memory layer for LLM agents.

Key updates:

pip install hmlr Drop-in LangGraph memory node (example agent included) New dossier system for reconstructing causal chains from long-term memory

It passes Hydra9 Hard Mode, a 20+ turn multi-hop test with 9 aliases, 8 policy revocations, and all turns in isolation. On the final query, no prior turns are in context, answer comes purely from long-term memory. Full reproducible test LangGraph memory node drop-in available from pip install or repo.