Graphify: AI Skill Turns Codebases into Knowledge Graphs

Graphify revolutionizes codebase understanding by transforming folders of code, docs, papers, images, and videos into interactive knowledge graphs. Type /graphify in Claude Code, Cursor, Aider, OpenClaw, or 12+ AI assistants to instantly extract architecture, god nodes, surprising connections, and design rationale. 71.5x token reduction vs raw files, supports 25+ languages via tree-sitter AST, local Whisper transcription for videos. Persistent graphs, git hooks, always-on assistant integration—perfect for navigating complex projects without grep hell.

Graphify: Transform Any Codebase into a Queryable Knowledge Graph

Imagine dropping a complex codebase, research papers, screenshots, and meeting recordings into a folder, then instantly getting a complete architectural map with god nodes, surprising connections, and design rationale. That's Graphify—the AI coding assistant skill that turns chaos into clarity.

What Makes Graphify Different?

Most AI assistants grep through files linearly. Graphify builds a persistent knowledge graph that captures:

  • AST-extracted structure (classes, functions, imports, call graphs) from 25+ languages via tree-sitter
  • Semantic relationships across files (two functions solving same problem without calling each other)
  • Design rationale from docstrings, #WHY: comments, and docs
  • Multimodal extraction from PDFs, images, videos (local Whisper transcription)
  • Confidence scoring on every inferred relationship

Result: 71.5x fewer tokens per query vs reading raw files. Your assistant navigates by structure, not keywords.

Slash Command Magic

Type /graphify . in any of these platforms:

Platform Command
Claude Code /graphify
Cursor /graphify
Aider /graphify
OpenClaw /graphify
GitHub Copilot CLI /graphify
Trae, Kiro, Gemini CLI, more...

Output:

graphify-out/
├── graph.html (interactive vis.js graph)
├── GRAPH_REPORT.md (god nodes + insights)
├── graph.json (persistent, queryable)
└── cache/ (SHA256 incremental updates)

Always-On Intelligence

Run graphify claude install (or platform equivalent) and your assistant automatically reads the graph summary before every file search:

"Knowledge graph exists. Read GRAPH_REPORT.md for god nodes and communities before grepping."

No more missing architectural context. Perfect for PR reviews, onboarding, debugging.

Production Features

/graphify . --watch          # Auto-sync as files change
graphify hook install        # Git hooks rebuild on commit
graphify query "auth flow"   # Graph-powered deep queries
/graphify add paper.pdf      # Incremental knowledge growth

Outputs: Obsidian vaults, Neo4j cypher, GraphML, SVG, agent-crawlable wikis.

Privacy-First

  • ✅ Code AST parsing: 100% local (tree-sitter)
  • ✅ Video transcription: 100% local (faster-whisper)
  • ✅ No telemetry/analytics
  • ✅ Only sends docs/images to your existing AI provider

Real-World Impact

Karpathy repos + papers + images: 52 files → 71.5x token compression httpx library: Full call graph + docstring rationale Transformers paper + code: Cross-modal concept linking

[Star 26.1k | 2.8k forks] on GitHub. MIT license. pip install graphifyy.

Try it: Clone the worked examples and run /graphify yourself. Your AI coding workflow will never be the same.