safishamsi/graphify

⭐ 45,847  ·  Python  ·  GitHub Repo

AI coding assistant skill (Claude Code, Codex, OpenCode, Cursor, Gemini CLI, and more). Turn any folder of code, SQL schemas, R scripts, shell scripts, docs, papers, images, or videos into a queryable knowledge graph. App code + database schema + infrastructure in one graph.

antigravity claude-code codex gemini graphrag knowledge-graph leiden openclaw

1-Sentence Summary

Turn any folder of code, docs, and images into a queryable knowledge graph with honest edge tagging.

🔥 Key Capabilities & USP

• Multimodal Knowledge Graph Construction – Ingests code (via tree-sitter AST), PDFs, markdown, images, screenshots, and diagrams into a single unified graph. Solves the pain point of navigating large, mixed-format codebases where relationships between code, documentation, and design artifacts are invisible.
• Honest Edge Tagging System – Every relationship is tagged as EXTRACTED, INFERRED, or AMBIGUOUS, distinguishing found connections from guessed ones. This is a standout differentiator that gives engineers confidence in query results and prevents blind trust in speculative links.
• Persistent Graph & Interactive Outputs – Generates an interactive HTML graph (vis.js), an Obsidian vault, Wikipedia-style wiki articles, and a persistent graph.json. The graph can be queried weeks later without re-reading files, delivering up to 71.5x token reduction per query compared to raw file reading.
• Smart Incremental Updates – SHA256 caching ensures re-runs only process changed files; --update merges new extractions into the existing graph; --watch auto-syncs as files change. This eliminates the pain of full rebuilds and keeps your knowledge base current with zero manual overhead.
• Git Integration – graphify hook install adds a post-commit hook that automatically rebuilds the graph after every commit. Your knowledge graph stays synchronized with your version control workflow without background processes.

Technical Architecture

Component	Technology	Role
Graph Engine	NetworkX + Leiden clustering (graspologic)	Core graph construction and community detection
Code Parsing	tree-sitter	AST extraction for code files
LLM Extraction	Claude	Concept/relationship extraction from docs, images, and non-code artifacts
Visualization	vis.js	Interactive HTML graph output
Caching	SHA256 hashing	Incremental update detection and change tracking
Architecture	Fully local, no server required	Zero infrastructure dependencies; outputs include graph.html, obsidian/ vault, wiki/ articles, GRAPH_REPORT.md, graph.json, and cache/ directory

Quick Start Guide

pip install graphifyy && graphify install

mkdir -p ~/.claude/skills/graphify
curl -fsSL https://raw.githubusercontent.com/safishamsi/graphify/v1/skills/graphify/skill.md \
  > ~/.claude/skills/graphify/SKILL.md

/graphify .                        # run on current directory
/graphify ./raw                    # run on a specific folder
/graphify ./raw --mode deep        # more aggressive INFERRED edge extraction
/graphify ./raw --update           # re-extract only changed files, merge into existing graph
/graphify add https://arxiv.org/abs/1706.03762        # fetch a paper, save, update graph
/graphify add https://x.com/karpathy/status/...       # fetch a tweet
/graphify query "what connects attention to the optimizer?"
/graphify path "DigestAuth" "Response"
/graphify explain "SwinTransformer"
/graphify ./raw --watch            # auto-sync graph as files change
/graphify ./raw --wiki             # build agent-crawlable wiki
/graphify ./raw --svg              # export graph.svg
/graphify ./raw --graphml          # export graph.graphml (Gephi, yEd)
/graphify ./raw --neo4j            # generate cypher.txt for Neo4j
/graphify ./raw --mcp              # start MCP stdio server
graphify hook install              # post-commit git hook

Pros, Cons & Use Cases

Pros

• Dramatic token reduction – Up to 71.5x on mixed corpora of 50+ files, making agent queries radically cheaper and faster
• Fully multimodal – Code, docs, images, PDFs, screenshots, and whiteboard photos all in one graph
• Persistent across sessions – graph.json can be queried weeks later without re-reading files
• Honest edge tagging – Distinguishes EXTRACTED (found) vs INFERRED (guessed) vs AMBIGUOUS connections
• Multiple export formats – HTML, Obsidian, Wiki, Neo4j, SVG, GraphML, MCP server

Cons

• Requires Claude Code and Python 3.10+ – Not a standalone tool; depends on the Claude ecosystem
• LLM re-pass needed for doc/image changes – Only code AST rebuilds are instant; document and image changes require a full LLM extraction pass
• Token reduction is minimal (~1x) for very small corpora – The value proposition weakens for projects with fewer than ~10 files
• PyPI package temporarily named graphifyy – Due to a naming conflict, the install command uses a non-intuitive package name

Who should NOT use this?

• Developers working on small, single-language projects (e.g., a 6-file Python script) – The overhead of graph construction isn't justified by the minimal token savings
• Teams without access to Claude – The tool is tightly coupled to Claude for LLM extraction; no alternative LLM backend is mentioned
• Users needing real-time, streaming updates – While --watch exists, doc/image changes require a non-instant LLM pass
• Projects with zero documentation or non-code artifacts – The multimodal value is lost if you only have code

Ideal Use Cases

• Large, multi-language monorepos – Discover hidden connections between frontend code, backend services, database schemas, and infrastructure configs
• Research repositories – Combine papers (PDFs), code implementations, experiment notes, and whiteboard photos into a single queryable graph
• Onboarding new team members – Let new hires query "what connects authentication to the payment pipeline?" and get an instant, tagged relationship map
• Legacy codebase archaeology – Run on a folder of undocumented code and documentation to surface implicit architecture and dependencies
• AI agent knowledge bases – Build persistent, agent-crawlable wikis that reduce token costs by 71x per query

Community & Activity

With 45,847 stars and a last update on May 10, 2026, Graphify has clearly struck a nerve in the developer community. This is not a dormant project—it's actively maintained and rapidly gaining traction. The topic tags (antigravity, claude-code, codex, gemini, graphrag, knowledge-graph, leiden, openclaw, rag, skills, tree-sitter) show a project that's positioning itself at the intersection of knowledge graphs, RAG, and AI coding assistants. The momentum suggests a tool that's solving a real, painful problem for developers working with heterogeneous codebases. If you're evaluating this for your team, you're looking at a project with strong community validation and active development.