What Is Graphify? The Open-Source Tool That Gives AI Assistants Codebase Memory

AI Tools · Developer Productivity · Open Source · Knowledge Graphs

AI assistants are brilliant at reading a single file. They are terrible at understanding how 200 files connect. Graphify fixes that. And it went viral overnight.

Auriga IT Editorial Team June 2026 14 min read Tags: Claude Code · Knowledge Graphs · Open Source

01The problem every developer knows

You are onboarding to a large codebase. Or you have been on the project for months but the architecture is sprawling. Microservices, multiple databases, API layers, scattered documentation. You open your AI coding assistant, ask how the authentication module connects to the user data layer, and get a file-by-file scan that burns thousands of tokens and still misses the structural picture.

This is not a model intelligence problem. It is a context problem.

AI coding assistants like Claude Code, Codex, Cursor, and Gemini CLI operate on flat-file context. They have no map of how your codebase is actually organised, how functions call each other across modules, why certain architectural decisions were made, or how a Figma doc relates to the API handler it describes. Every single query, they start from scratch.

02What is Graphify?

Graphify is an open-source AI coding assistant skill. It is a slash command (/graphify) you invoke inside AI coding assistants like Claude Code, OpenAI Codex, Cursor, Gemini CLI, OpenCode, Aider, Kiro, and more.

When you run it, it reads every file in your project folder and builds a queryable knowledge graph showing the structure, relationships, and reasoning behind everything in your project. Instead of your AI assistant grepping through files on every query, it navigates a compact, pre-built map.

03The origin story: born in 48 hours

On April 1, 2026, Andrej Karpathy, former Tesla AI director and OpenAI co-founder, posted on X describing a workflow he wished existed. A way to drop papers, tweets, screenshots, and code notes into a folder and actually be able to query them meaningfully later, with an AI that understood the relationships between all of it.

Within 48 hours, Safi Shamsi published Graphify on GitHub. Shamsi holds an MSc in Data Science with Distinction from the University of Birmingham, where his thesis focused specifically on knowledge-graph-based hybrid RAG systems for academic search. He had the exact technical background to turn Karpathy's vision into working software immediately.

The community response was immediate. A tweet announcing the project earned over 12,000 likes, framed simply as: "Karpathy asked for LLM knowledge graphs, and someone built it." Within 10 days, Graphify had 22,000 GitHub stars. By June 2026 it had crossed 58,300 stars and 1.2 million PyPI downloads.

04How Graphify works: the 3-pass architecture

05What you get as output

After running /graphify . in your project folder, three files land in graphify-out/:

06How to install and set up Graphify

07Which AI coding assistants does Graphify support?

As of June 2026, Graphify works as a registered skill across 10+ AI coding platforms:

For Claude Code and Gemini CLI, Graphify installs a hook that fires automatically before search-style tool calls, nudging your assistant toward graph queries instead of raw file reads. For other platforms, it writes persistent instruction files (AGENTS.md, .cursor/rules/) that achieve the same behaviour.

08Is your code safe?

This is the question every security-conscious developer should ask before using any tool that touches source code. Here is the full breakdown.

09Who should use Graphify?

10What actually changes when you use it

The clearest way to describe the difference: without Graphify, your AI assistant is a very fast reader. With Graphify, it becomes a senior engineer who already knows the codebase.

The difference is most visible on structural questions like these:

This kind of structural AI reasoning is exactly the capability that separates surface-level AI adoption from real AI-driven engineering. At Auriga IT, we apply this thinking across products for clients like Ferns N Petals and Yes Bank, building data and AI layers that give software systems genuine contextual intelligence.

11How Auriga IT can help your organisation

Graphify is a strong tool for individual developers and small teams. But for organisations looking to embed AI deeply into engineering workflows, across large codebases, multi-team environments, and enterprise-grade products, there is more ground to cover than a single open-source skill can address.

At Auriga IT, we work with startups, enterprises, and product companies to design and implement AI-assisted development workflows, build knowledge graph and RAG architectures, modernise legacy codebases so AI tools have clean systems to work with, and deliver Data, AI, and Analytics products from initial assessment through to production deployment.

We have delivered 1,000+ projects across e-commerce, fintech, healthcare, and infrastructure. If you are evaluating how AI tooling fits into your engineering roadmap, talk to our team for a no-commitment initial conversation.

12The road ahead: from Graphify to Penpax

The Graphify project, maintained by Safi Shamsi and now 71 contributors with releases roughly every other day, is moving toward Penpax: an always-on layer that applies the same graph approach to your entire working life. Meetings, browser history, emails, files, and code, updated continuously in the background.

The underlying insight is powerful. The knowledge graph is not just a codebase tool. It is a general-purpose memory architecture for AI agents. As agentic AI systems become more capable, the challenge of what they remember and how they navigate it becomes the critical constraint. Graphify is an early, well-executed answer to that challenge.

13Quick reference

14Frequently asked questions