graphify-rs

AI-powered knowledge graph builder

Transform code, docs, papers, and images into queryable, interactive knowledge graphs.

中文文档 | CLI Reference | Architecture | Changelog

---

Why graphify-rs?

Built around Andrej Karpathy's /raw folder workflow: drop anything into a folder — papers, tweets, screenshots, code, notes — and get a structured knowledge graph that shows you what you didn't know was connected.

Three things it does that an LLM alone cannot:

	Feature	Why it matters
1	Persistent graph	Relationships survive across sessions. Query weeks later without re-reading.
2	Honest audit trail	Every edge tagged EXTRACTED, INFERRED, or AMBIGUOUS. Facts vs. guesses, always clear.
3	Cross-document surprise	Community detection finds connections you'd never think to ask about.

Quick Start

# Install
cargo install graphify-rs

# Build a knowledge graph (free, fast, no API key needed)
graphify-rs build --no-llm

# Explore interactively
open graphify-out/graph.html         # macOS
# xdg-open graphify-out/graph.html   # Linux

# Query the graph
graphify-rs query "how does auth work?"

# (Optional) Add semantic extraction via Claude API
export ANTHROPIC_API_KEY=sk-...
graphify-rs build

Performance

Rust rewrite of graphify (Python) — fully compatible graph.json output.

	Python	Rust
Speed	~204ms	~24ms (8.5x faster)
Memory	~48MB	~1MB (48x less)
AST parsing	Regex only	11 native tree-sitter + regex fallback
Community detection	Louvain	Leiden (with refinement)
MCP server	-	15 tools over JSON-RPC 2.0
Export formats	7	9 (+ Obsidian, split HTML)
Extraction	Sequential	Parallel (rayon, configurable -j)

How It Works

 Source Files              graphify-rs build
 ┌──────────┐    ┌──────────────────────────────────────────────────────┐
 │ .py .rs  │    │                                                      │
 │ .go .ts  │───>│  detect -> extract -> build -> cluster -> analyze -> export
 │ .md .pdf │    │                                                      │
 └──────────┘    └──────────┬───────────────────────────────────────────┘
                            v
                  graphify-out/
                  ├── graph.json          queryable graph data
                  ├── graph.html          interactive visualization
                  ├── GRAPH_REPORT.md     analysis report
                  ├── wiki/               per-community wiki pages
                  └── obsidian/           Obsidian vault

Pass 1 — AST extraction (free, always runs): tree-sitter parses 21 languages into functions, classes, imports, calls. All edges tagged EXTRACTED (confidence 1.0).

Pass 2 — Semantic extraction (optional, --no-llm to skip): Claude API discovers conceptual links, shared assumptions, design rationale. Edges tagged INFERRED (confidence 0.4–0.9).

Graph Algorithms

7 advanced algorithms beyond basic traversal:

Algorithm	What it does
Leiden clustering	Community detection with internal connectivity guarantee
PageRank	Structural importance (not just degree) — finds true architectural pillars
Tarjan's SCC	Dependency cycle detection — surfaces circular imports
Dijkstra weighted path	Shortest path weighted by edge confidence
Node2Vec embedding	Graph similarity search — finds redundant/refactorable code
Incremental clustering	Re-clusters only changed communities on rebuild
Smart summarization	Three-level abstraction (detailed → community → architecture) for LLM token budgets

See docs/ARCHITECTURE.md for complexity analysis.

Supported Languages (21)

Native tree-sitter	Regex fallback
Python, JavaScript, TypeScript, Rust, Go, Java, C, C++, Ruby, C#, Dart	Kotlin, Scala, PHP, Swift, Lua, Zig, PowerShell, Elixir, Obj-C, Julia

Agent Integration

graphify-rs install              # install skill for AI coding agents
graphify-rs serve                # start MCP server (15 tools)

Agents auto-check the graph before architecture questions and rebuild after code changes. Works with Claude Code, Codex, OpenCode, and more.

15 MCP tools: query_graph, pagerank, detect_cycles, smart_summary, find_similar, shortest_path, and 9 more.

Architecture

14-crate Cargo workspace — see docs/ARCHITECTURE.md for the full design.

Crate	Role
graphify-core	Data models, graph structure, confidence system
graphify-extract	AST extraction (21 languages), Claude API semantic extraction
graphify-cluster	Leiden community detection, incremental re-clustering
graphify-analyze	PageRank, cycles, embeddings, god nodes, temporal risk
graphify-serve	MCP server (15 tools), smart summarization
graphify-export	9 formats: JSON, HTML, SVG, GraphML, Cypher, Wiki, Obsidian, Report
+ 8 more	Cache, security, ingestion, watch, hooks, benchmark, detect, build

Output Formats

File	Description
graph.json	NetworkX-compatible node_link_data JSON
graph.html	Interactive vis.js visualization (dark theme, auto-pruning)
html/	Per-community HTML pages with navigation
GRAPH_REPORT.md	God nodes, surprising connections, suggested questions
graph.svg / graph.graphml	Static visualization / graph editor import
cypher.txt	Neo4j import script
wiki/ / obsidian/	Wiki pages / Obsidian vault with wikilinks

CLI at a Glance

graphify-rs build [--path .] [--no-llm] [--format json,html]   # build graph
graphify-rs query "question" [--dfs] [--budget 2000]            # query
graphify-rs watch --path .                                       # auto-rebuild
graphify-rs serve                                                 # MCP server
graphify-rs diff old.json new.json                               # compare
graphify-rs stats graph.json                                     # statistics

Full reference: docs/CLI.md (21 subcommands)

Contributing

See CONTRIBUTING.md for setup, code style, testing, and PR guidelines.

License

MIT — see LICENSE.

Rust rewrite of graphify by safishamsi.