knowledge-graph

Run the graphify pipeline on the target folder: it detects files, extracts code structure via AST without an LLM, extracts semantic entities from docs and images, then clusters the merged graph. Outputs include graph.html, graph.json, and GRAPH_REPORT.md.

Supported inputs include code files (.py, .ts, .go, and others), documents (.md, .txt), papers (.pdf), images, and video/audio files. Video and audio are transcribed with Whisper and treated as documents during extraction.

No external API key is required. Code is extracted structurally via AST with no LLM, and semantic extraction for docs, papers, and images uses the agent's own model. A code-only corpus skips semantic extraction entirely.

Use the --update flag for incremental re-extraction of only new or changed files, or --watch to auto-rebuild on changes. A git post-commit hook can also rebuild the graph automatically after every commit.

Yes. The --neo4j flag generates a Cypher file and --neo4j-push writes directly to a running Neo4j instance using MERGE. The --obsidian flag generates an Obsidian vault with one note per node plus a canvas layout.

An empty graph error means extraction produced no nodes, typically because all files were skipped, the corpus contained only unsupported binaries, or extraction failed. Check the detect step output for skipped or unsupported files and re-run on a valid path.