KHMS is a long-term memory system for LLM agents built from immutable markdown cards in a git repository, extracted from a single-operator deployment running daily since mid-2026. Cards carry YAML frontmatter with evidence levels and provenance; corrections supersede rather than edit. The system includes hook-driven recall and a propose-review-approve pipeline, with documentation for wiring it into Claude Code. Its storage format closely parallels Google Cloud's Open Knowledge Format, which lacks KHMS's epistemic layer.
No score is assigned. Sources and their independence are shown in the citation chain below.