This bridges Claude Code's auto-memory markdown files into AgentDB with proper semantic search. It reads from `~/.claude/projects/*/memory/*.md`, generates 384-dim ONNX embeddings, and stores them in AgentDB's HNSW index with deduplication at 0.95 cosine similarity. Runs automatically on session start, but you can manually import all projects or force a refresh. The real win is unified search across all your session memories instead of being stuck in per-project silos. If you have ruflo-ruvector loaded, it also enables hybrid search and graph RAG across memory entries. Basically turns scattered memory files into a queryable knowledge base.
npx skills add https://github.com/ruvnet/ruflo --skill memory-bridge