graphify

A Go implementation of graphify — turn source code into a knowledge graph, cluster it into communities, and generate a visual report.

What it does

1. Detect source files in a project directory
2. Extract nodes (functions, classes, modules) and edges (calls, inherits, imports) using tree-sitter AST parsing
3. Build a knowledge graph from the extracted entities
4. Cluster the graph into communities using the Louvain algorithm
5. Analyze the graph for god nodes and surprising connections
6. Export an interactive HTML visualization, JSON graph, and Markdown report

Supported languages (24)

C, C++, C#, Dart, Elixir, Elm, Go, Haskell, Java, JavaScript/TypeScript (JSX/TSX), Julia, Kotlin, Lua, Objective-C, PHP, PowerShell, Python, R, Ruby, Rust, Scala, Svelte, Swift, Vue, Zig

Installation

go install github.com/sjhorn/graphify/cmd/graphify@latest

Usage

# Analyze a project directory
graphify /path/to/project

# Specify output directory
graphify -out results /path/to/project

# Verbose output
graphify -verbose /path/to/project

# Add graphify prompt to CLAUDE.md
graphify claude
graphify claude /path/to/project

# Add graphify prompt to AGENTS.md
graphify agents
graphify agents /path/to/project

Output

All output is written to the output directory (graphify-out/ by default):

• graph.json — full graph data with nodes, edges, and community assignments
• graph.html — interactive force-directed visualization
• GRAPH_REPORT.md — Markdown report with community summaries, god nodes, and surprising connections
• cache/ — per-file extraction cache (SHA256-keyed), automatically used on subsequent runs

Caching

Extraction results are cached per file based on content hash (SHA256). On subsequent runs, unchanged files are loaded from cache instead of being re-extracted. For Markdown files, only the body is hashed — frontmatter changes (e.g. reviewed dates) don't invalidate the cache.

Use -verbose to see cache hit/miss stats.

How extraction works

Each source file is parsed into a tree-sitter AST. Language-specific extractors walk the AST to identify:

• Nodes: files, classes, functions, methods, interfaces, enums, modules
• Edges with relation types:
  • contains — file contains a class, class contains a method
  • method — class has a method
  • calls — function calls another function
  • inherits — class extends or implements another
  • imports — file imports a module or symbol
  • case_of — enum variant belongs to an enum

Attribution

This is a Go rewrite of graphify by Safi Shamsi.

Key dependencies:

• tree-sitter/go-tree-sitter — AST parsing
• gonum — graph algorithms and community detection

License

MIT