graphify

Run /graphify with a target path to extract entities and relationships from code, docs, papers, and images into a single graph. It produces an interactive HTML visualization, a GraphRAG-ready JSON file, and a plain-language report with community labels.

Code files go through deterministic AST extraction that captures imports and structural relationships for free. Parallel LLM subagents then add semantic edges like call relationships, shared data structures, and architecture patterns that AST cannot detect.

Yes, video and audio files are transcribed to text using Whisper before extraction, with a domain hint derived from the corpus to improve accuracy. The transcripts are then treated as documents, and you can select a larger Whisper model with the --whisper-model flag.

Yes, the --neo4j flag generates a Cypher import file and --neo4j-push writes directly to a running Neo4j instance. The --obsidian flag generates a vault with one note per node plus a canvas file for structured community layout.

Use the --update flag for incremental runs that re-extract only new or changed files, since extraction results are cached per file. The --watch flag monitors a folder and rebuilds automatically on code changes without needing an LLM.

Corpora over 2 million words or 200 files trigger a warning and require selecting a subfolder first. Graphs with more than 5000 nodes skip HTML visualization and should be explored through the Obsidian vault export instead.