Graphify — Knowledge Graph for AI Coding Assistants
Graphify maps entire codebases into queryable knowledge graphs using tree-sitter AST parsing, saving AI agents from reading raw files.
AI coding assistants like Claude Code, Cursor, and Codex are powerful — but they have a fundamental limitation: they need to read raw source files to understand a codebase. For large projects, this burns through context windows and API credits at an alarming rate.
Graphify solves this problem by mapping your entire project — code, docs, PDFs, images, and more — into a structured, queryable knowledge graph that AI assistants can navigate instantly.
What is Graphify?#
Graphify is an open-source, local-first tool that parses a folder into a real node-and-edge graph, stored as graph.json. AI coding assistants query this graph instead of reading raw files, drastically reducing token usage while preserving structural accuracy.
It supports 36+ programming languages via tree-sitter AST parsing, handles multimodal content (PDFs, images, videos, spreadsheets), and produces three key artifacts in graphify-out/:
graph.html— Interactive, clickable graph visualizationGRAPH_REPORT.md— Architectural summary with key concepts and cross-file connections (Obsidian-compatible)graph.json— Raw knowledge graph for programmatic querying
The project has rapidly gained traction, accumulating close to 100,000 GitHub stars within months of its launch.
How It Works#
Graphify uses a two-pass hybrid processing architecture:
Pass 1: Code (Deterministic AST Parsing)#
Local tree-sitter parsing across 36+ languages. Completely offline — zero LLM calls, nothing leaves your machine. Extracts:
- Classes, functions, methods, interfaces
- Import/export relationships
- Call graphs and inheritance hierarchies
- Variable references and type information
Pass 2: Docs & Media (Semantic Extraction)#
For non-code files — PDFs, images, audio, video, spreadsheets — Graphify uses your configured AI backend (Claude, OpenAI, Gemini, Ollama, Bedrock, etc.) to extract semantic concepts and link them to the code graph.
Edge Labeling#
Every relationship is explicitly tagged:
EXTRACTED— Explicitly present in the source (e.g., a function call)INFERRED— Derived by Graphify’s resolution algorithms (e.g., conceptual similarity)
This lets developers and AI agents distinguish verified relationships from algorithmic inferences.
Community Detection#
Graphify applies Leiden clustering (LLM-free) to segment code into logical subsystems and identify “God nodes” — the most central, heavily-connected components in your codebase.
Supported Data Sources#
| Type | Formats |
|---|---|
| Code | Python, TypeScript, JavaScript, Go, Rust, Java, C/C++, C#, Ruby, Swift, Kotlin, Scala, PHP, Lua, Zig, SQL, Terraform/HCL, and more (36+ grammars) |
| Documentation | Markdown, MDX, HTML, reStructuredText, YAML, JSON, Obsidian wikilinks |
| Office | Word (.docx), Excel (.xlsx), Google Workspace |
| Media | PDFs, PNG/JPG/WebP/GIF, audio/video (.mp4, .mov, .mp3, .wav), YouTube URLs, arXiv papers |
Installation#
# Prerequisites: Python 3.10+
# Install the CLI (package name: graphifyy)
uv tool install graphifyy
# or
pipx install graphifyy
# Register the AI skill
graphify install
# Run against a project
graphify .To scope the skill to a specific project:
graphify install --projectKey Features#
Strict Mode#
Enforceable via Claude Code integration to block the first raw source file read and force the agent to consult the graph first.
Git Hook Integration#
Auto-rebuild code graphs on commit/checkout:
graphify hook installQuery Engine#
AI agents run targeted subgraph queries instead of reading entire files:
graphify query "How does authentication work?"
graphify path "auth-service" "user-service"Token Optimization#
By feeding AI assistants precise subgraphs instead of raw files, Graphify drastically lowers token usage and API costs — especially valuable for large codebases.
Graphify vs. Alternatives#
| Feature | Graphify | Vector RAG (mem0, etc.) | Grepping |
|---|---|---|---|
| Structural accuracy | ✅ AST-level precision | ❌ Similarity-based | ✅ Exact match |
| LLM credits to index | ❌ Zero (Pass 1 is local) | ⚠️ High embedding costs | ✅ Zero |
| Multimodal support | ✅ Code + docs + media | ✅ Text only | ❌ Text only |
| Relationship types | ✅ EXTRACTED vs INFERRED | ❌ Semantic similarity only | ❌ None |
| Interactive graph | ✅ graph.html | ❌ No | ❌ No |
| Zero-shot recall@10 | ~0.497 | Variable | N/A |
Use Cases#
Onboarding to large codebases — New team members explore GRAPH_REPORT.md and query the graph instead of spending days reading files.
Code review impact analysis — Track how changes propagate across modules before making edits.
Multimodal project documentation — Merge codebase logic with architecture diagrams, meeting PDFs, and whiteboard photos into one unified knowledge graph.
Cost-efficient CI/CD — Feed AI code review agents precise subgraphs instead of entire repositories.
Summary#
Graphify transforms how AI coding assistants understand code. Instead of dumping raw files into context windows — expensive, wasteful, and imprecise — it builds a structured knowledge graph that agents can query with surgical precision.
With zero-cost code indexing (tree-sitter, local-only), broad multimodal support, and visual exploration tools, Graphify has rapidly become an essential tool in the AI-assisted development stack.