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

Graphify turns any folder into a queryable knowledge graph, slashing LLM token costs by 71.5x.

🔥 Key Capabilities & USP

• Multimodal Knowledge Graph Construction: Ingests code (AST/tree-sitter), SQL schemas, PDFs, images, and whiteboard photos into a single, unified graph. Solves the pain of fragmented knowledge across mixed-format codebases and documentation.
• 71.5x Token Reduction: By building a persistent graph, queries against the graph use dramatically fewer tokens than re-reading raw files. This is the core USP—directly reduces AI coding assistant costs and latency.
• Transparent Edge Tagging: Every relationship is tagged as EXTRACTED, INFERRED, or AMBIGUOUS. You always know whether a connection was found in source code or guessed by the LLM, eliminating the "black box" trust issue.
• Auto-Sync & Git Integration: --watch mode rebuilds the graph in real-time as files change. The graphify hook install command adds a post-commit git hook, so your knowledge graph is always up-to-date with zero manual effort.
• Rich Export Ecosystem: Outputs to interactive HTML, Obsidian vaults, Wikipedia-style wikis, SVG, GraphML (Gephi/yEd), Neo4j Cypher, and MCP stdio server—making the graph usable across visualization, analysis, and agentic workflows.

Technical Architecture

Component	Technology	Role
Graph Engine	NetworkX	In-memory graph representation and traversal
Clustering	Leiden (via graspologic)	Community detection for graph structure
Code Parsing	tree-sitter	AST-based extraction of code concepts and relationships
LLM Extraction	Claude (Anthropic)	Concept and relationship extraction from docs, images, PDFs
Visualization	vis.js	Interactive HTML graph rendering
Caching	SHA256	Only reprocesses changed files on re-runs
Storage	Local filesystem	No Neo4j or server dependencies; fully local operation

Architectural Highlight: Runs entirely locally with zero external infrastructure. The --update flag merges new extractions into an existing graph without rebuilding from scratch, enabling incremental growth.

Quick Start Guide

# Install (note: PyPI name is graphifyy due to naming conflict)
pip install graphifyy && graphify install

# Run on current directory
/graphify .

# Run on a specific folder with deep inference
/graphify ./raw --mode deep

# Re-extract only changed files, merge into existing graph
/graphify ./raw --update

# Fetch a paper and update the graph
/graphify add https://arxiv.org/abs/1706.03762

# Query the graph
/graphify query "what connects attention to the optimizer?"
/graphify path "DigestAuth" "Response"
/graphify explain "SwinTransformer"

# Auto-sync as files change
/graphify ./raw --watch

# Export to various formats
/graphify ./raw --wiki
/graphify ./raw --svg
/graphify ./raw --graphml
/graphify ./raw --neo4j
/graphify ./raw --mcp

# Install post-commit git hook
graphify hook install

Pros, Cons & Use Cases

Pros

• Dramatic token reduction (71.5x) directly lowers costs for AI coding assistant users.
• Fully local—no Neo4j, no cloud dependencies, no data leaving your machine.
• Transparent extraction with edge tagging builds trust in the knowledge graph.
• Multimodal support (code, docs, images, PDFs) covers real-world heterogeneous repositories.
• Rich export formats integrate with existing tools (Obsidian, Gephi, Neo4j, MCP).

Cons

• Requires Python 3.10+ and Claude Code for LLM-based extraction—not a standalone tool.
• LLM extraction latency for docs/images may slow initial graph construction on large corpora.
• PyPI naming conflict means install command is pip install graphifyy instead of graphify.
• No built-in vector search—queries rely on graph traversal, not semantic similarity.

Who should NOT use this?

• Developers working with small, single-language codebases—the overhead of graph construction outweighs benefits for projects you can already navigate manually.
• Teams without access to Claude Code—LLM extraction is a core feature; without it, you lose doc/image support and inference capabilities.
• Users needing real-time collaborative graphs—Graphify is a local, single-user tool with no multi-user or server mode.
• Projects requiring strict air-gapped operation—while the tool runs locally, Claude Code requires API access for LLM features.

Ideal Use Cases

• Large, mixed-format repositories (code + SQL + docs + images) where navigating manually is painful.
• AI coding assistant workflows (Claude Code, Codex, Cursor, Gemini CLI) where token costs are a concern.
• Research codebases with papers, whiteboard photos, and experimental scripts that need to be connected.
• Legacy system onboarding—quickly build a queryable map of undocumented code and infrastructure.
• Continuous integration pipelines—use the git hook to keep a live knowledge graph synced with every commit.

Community & Activity

With 45,847 stars, Graphify has clearly struck a nerve in the developer community. This is not a niche experiment—it's a rapidly adopted tool for anyone working with AI coding assistants on complex projects. The project was last updated on May 10, 2026, indicating active maintenance and a responsive maintainer. The breadth of topics (antigravity, claude-code, codex, gemini, graphrag, knowledge-graph, leiden, openclaw, rag, skills, tree-sitter) shows a project that's thinking about the full ecosystem, not just a single use case. If you're building agentic workflows or managing heterogeneous codebases, this is a project worth watching—and contributing to.