Embeddings

Embedding engine commands:

1. Call mcp__plugin_ruflo-core_ruflo__embeddings_status to check the ONNX embedding engine.
2. Show: model (all-MiniLM-L6-v2), dimensions (384), HNSW index status, cache hit rate.
3. If not initialized, suggest calling mcp__plugin_ruflo-core_ruflo__embeddings_init.
4. For search, use mcp__plugin_ruflo-core_ruflo__embeddings_search with the user's query (namespace-filtered).
5. For large-corpus search with memory pressure, use the RaBitQ quantized path (32× memory reduction):
   • mcp__plugin_ruflo-core_ruflo__embeddings_rabitq_build (one-time, after corpus is loaded)
   • mcp__plugin_ruflo-core_ruflo__embeddings_rabitq_search (Hamming-prefilter to top-N candidates)
   • mcp__plugin_ruflo-core_ruflo__embeddings_rabitq_status for index health
6. For hierarchical data (taxonomies, code trees), use mcp__plugin_ruflo-core_ruflo__embeddings_hyperbolic (Poincare ball model).
7. The substrate-level entry point mcp__plugin_ruflo-core_ruflo__embeddings_neural exists; in normal use it's covered by embeddings_init + embeddings_generate.