slm-build-graph

Build a knowledge graph from memories by extracting entities with TF-IDF and discovering relationships via Leiden clustering to improve search quality. Run the build command to create or refresh graph nodes from existing memory data.

Rebuild your memory knowledge graph when search quality is weak, after bulk memory imports, or to improve performance in memory-rich environments. Refreshing the graph updates entity nodes and semantic clusters for better retrieval accuracy.

You need python-igraph and leidenalg as optional dependencies to enable Leiden clustering for semantic topic grouping. Core entity extraction with TF-IDF and memory relationship discovery function without these libraries on SuperLocalMemory V3.

TF-IDF entity extraction identifies significant terms across memories to create graph nodes for memory search optimization. Shared entities and similarity metrics then link these nodes, revealing hidden connections that enhance retrieval relevance.

Cluster memories into semantic topics using the Leiden algorithm on a knowledge graph built from shared entities. This optional clustering groups related memories into distinct communities, improving navigability and contextual search performance.