Graphify is a Claude Code skill that turns a folder of mixed material into a knowledge graph. Invoked with /graphify, it reads code, markdown, PDFs, and images, extracts concepts and relationships, and connects them into a single graph you can browse, query, and reuse across sessions without re-reading the raw files.
Extraction is per file type: source code in a dozen languages is parsed into an AST with tree-sitter followed by a call-graph pass; documents and papers go through Claude for concept and relationship extraction, with citation mining for PDFs; and images such as screenshots, diagrams, and whiteboard photos are read with Claude vision. Every edge is tagged EXTRACTED, INFERRED, or AMBIGUOUS so the graph records what was found versus inferred.
Output lands in a graphify-out/ directory: an interactive HTML graph, an Obsidian vault, a GRAPH_REPORT.md listing god nodes, ranked surprising connections, and suggested questions, and a persistent graph.json. A SHA256 cache means re-runs process only changed files. Communities are detected with Leiden clustering over NetworkX; the whole pipeline runs locally with no graph database or server.
On a 52-file corpus of code, papers, and images, querying the graph used 71.5x fewer tokens than reading the raw files; on a 4-file corpus the reduction was 5.4x, and for 6 files roughly 1x. Each worked example ships its input files and output so the numbers can be reproduced.
Features
- Multimodal input: code (.py .ts .js .go .rs .java .c .cpp .rb .cs .kt .scala .php), docs (.md .txt .rst), PDFs, and images (.png .jpg .webp .gif)
- Graph queries: query, path, and explain subcommands answer questions, trace paths between nodes, and describe a concept
- Incremental updates: --update re-extracts only changed files and merges into the existing graph
- Auto-sync: --watch rebuilds instantly on code saves (AST only) and flags doc or image changes for an LLM re-pass
- Git hook: graphify hook install adds a post-commit hook that rebuilds the graph after every commit
- Remote sources: graphify add fetches a paper or tweet, saves it, and updates the graph
- Agent-crawlable wiki: --wiki writes Wikipedia-style markdown articles per community with an index.md entry point
- Exports: SVG, GraphML (Gephi, yEd), Neo4j Cypher, and an MCP stdio server via --mcp
- Provenance tags: each edge is marked extracted, inferred, or ambiguous
- Deep mode: --mode deep extracts inferred edges more aggressively
