Integrating Graphify: Building Knowledge Graphs from Codebases and Documentation for AI Coding Assistants

Graphify is an open-source tool that helps build a comprehensive Knowledge Graph from your codebase and project documentation, optimizing collaboration with AI coding assistants like Claude Code, Cursor, or Gemini CLI. Instead of forcing the AI to re-read hundreds of source files every time you ask a question, Graphify indexes structural relationships using AST (Abstract Syntax Tree), helping reduce token consumption by up to 71.5 times and improving response accuracy.

Graphify Integration: Building Knowledge Graphs from Codebase and Documentation for AI Coding Assistants — Operational Architecture: Building Knowledge Graphs for AI Coding Assistants

1. What is Graphify and Why Do You Need a Knowledge Graph for Your Codebase?

If you have ever asked an AI to explain or refactor a large project, you know the drill: with every prompt, the AI has to re-scan tens of thousands of lines of code just to figure out which module a function calls or where a class inherits from. This not only slows down response times, but also bloats the context window and racks up unnecessary token costs.

Graphify completely solves this problem using the “City Map” principle:

The old approach: The AI acts like a pedestrian walking down every single street to find the way (manually reading file by file).
The Graphify approach: Builds a complete satellite map of the entire city in advance. Whenever a question is asked, the AI simply consults the relationship map to go straight to the target location.
Offline AST parsing: Graphify leverages the tree-sitter library to perform static syntax analysis entirely locally on your machine, ensuring absolute security by never sending your source code over the internet.

REAL-WORLD PERFORMANCE

Real-world benchmarks show that using a Knowledge Graph reduces the number of tokens fed into the AI by an average of 71.5x per architectural query, while completely eliminating hallucinations regarding project structure.

2. System Requirements and 3-Step Installation Process

Graphify requires a computer with Python 3.10 or higher installed. To manage the environment without version conflicts or PATH environment variable errors, the most standard method is using the uv package manager:

Step 1: Install the uv package manager

Open Terminal (on macOS / Linux) and paste the following command:

uv (Mac/Linux)

curl -LsSf https://astral.sh/uv/install.sh | sh

For Windows PowerShell (run as Administrator):

uv (Windows)

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Step 2: Install Graphify CLI on your system

Once you have uv, install Graphify into an isolated tool environment with a single command:

Install Graphify

uv tool install "graphifyy"

Step 3: Register the Skill with Your AI Coding Assistant

Run the following command so Graphify can automatically detect and register the skill with Claude Code, Cursor, or Gemini CLI installed on your machine:

graphify install

The command above will automatically generate slash command definitions in each AI assistant’s configuration directory without requiring any manual setup.

3. Three Core Commands to Initialize and Query the Graph

After completing the installation, you can directly run these commands right inside the Claude Code or Cursor chat window:

Initialization Commands

# Analyze the entire current project directory
/graphify .

# Analyze only the source code directory (src)
/graphify ./src

# Comprehensive deep analysis (discovers more edges/connections)
/graphify . --mode deep

WINDOWS NOTE

In Windows PowerShell, a leading forward slash may be misinterpreted as a directory path. Omit the leading slash and enter graphify . instead of /graphify ..

Once the graph is initialized for the first time, you can ask the AI any question about your architecture flow—for example: “Which components call the authenticate() function in the auth module, and which database tables does it affect?” — The AI will query the knowledge graph to answer instantly.

4. Meaning and structure of the 3 output files in graphify-out

Whenever analysis is complete, Graphify creates a directory named graphify-out/ directly inside your project, containing 3 important files:

1. graph.html (Interactive Map)

A visual web page that lets you drag, zoom, and inspect clusters between modules across your project like a galactic network.

2. GRAPH_REPORT.md (Architecture Report)

A Markdown document compiling code quality metrics: dependency bottlenecks, god-class files, and architecture optimization recommendations.

3. graph.json (Raw Graph Data)

A structured JSON file containing the complete list of Nodes and Edges for AI assistants to load into memory and look up at lightning speed.

5. Team Setup and Advanced Extension Packages

If you work in a team, commit the graphify-out/ folder to your project’s Git repository. This way, any team member who clones the project instantly has access to the existing knowledge graph without wasting time rebuilding it from scratch.

GITIGNORE CONFIGURATION

Add the line graphify-out/cost.json to your project’s .gitignore file, as this file only tracks your personal token costs and does not need to be shared across the team.

For specialized, advanced needs (such as reading financial documents, connecting to databases, or sharing via MCP), install the following extension packages:

Extension Packages

# Read PDF, Word, and Excel documents:
uv tool install "graphifyy[documents]"

# Connect directly to a PostgreSQL database:
uv tool install "graphifyy[postgres]"

# Serve the graph via the MCP Server protocol:
uv tool install "graphifyy[mcp]"

# Install the full suite of all features:
uv tool install "graphifyy[all]"

6. Manual Installation Guide for Skills from GitHub

If your network environment blocks automated commands, you can manually create the skill file .claude/skills/graphify/SKILL.md in your project directory with the following standardized content:

Standard SKILL.md

---
name: graphify
description: "Use knowledge graphs to analyze codebase structure and project documentation"
---
# Guide to loading graph context:
1. Read the graphify-out/graph.json file to understand the list of Nodes and Edges.
2. When the user asks about architecture, look up the connecting edges before reading the source code in detail.

7. Common Errors and Troubleshooting

Below are the 5 most common scenarios encountered when integrating Graphify, along with quick troubleshooting guides:

Error: graphify: command not found after installing via pip
▼

Cause: The pip command often installs packages into the system’s Python directory without adding the executable path to the PATH environment variable.
Solution: Uninstall the pip package and switch to uv using the command: uv tool install graphifyy. The uv tool will handle the PATH environment variable automatically and seamlessly.

Graph has missing nodes or broken links after running –update
▼

Solution: When restructuring large directories or renaming files in bulk, use the force flag to run a comprehensive re-analysis using the command: /graphify . --force for a 100% fresh scan instead of an incremental update.

VRAM out-of-memory error when running with local Ollama
▼

Solution: Increase the context window size for the local model using an environment variable prior to running: GRAPHIFY_OLLAMA_NUM_CTX=8192 graphify export.

Skill version mismatch warning in Claude Code
▼

Solution: Upgrade Graphify to the latest version and refresh skill files using the command chain: uv tool upgrade graphifyy && graphify install.

Git merge conflict in graph.json when collaborating in a team
▼

Solution: Install Graphify’s smart merge hook using the command: graphify hook install. This command will automatically resolve JSON conflicts without corrupting the graph structure.

8. Frequently Asked Questions (FAQ)

Answers to questions about costs and methods for integrating Graphify into your software development workflow:

Does Graphify send my project’s source code to external servers?
▼

Absolutely not. The entire abstract syntax tree (AST) parsing process via tree-sitter runs 100% locally on your machine. Only when you use advanced summary extraction commands with an LLM will the summary text be sent to your API.

Which programming languages does Graphify support?
▼

Graphify provides full support for most popular languages: JavaScript, TypeScript, Python, Go, Rust, Java, C/C++, PHP, Ruby, C#, as well as document formats such as Markdown, JSON, and YAML.

How long does it take to build a graph for a project with millions of lines of code?
▼

Thanks to tree-sitter’s ultra-fast C-binding AST parsing mechanism, a project with 100,000 lines of code takes only about 15 to 30 seconds to complete the entire initial relationship graph.