VIDEO
HIGHLIGHT
Log InSign up
Graphify: Instant Knowledge Graph for Claude Code/Antigravity (FREE)
SummaryTranscriptChat

Graphify: Instant Knowledge Graph for Claude Code/Antigravity (FREE)

Understanding Claude Code and Graphify

Introduction to Claude Code

  • The speaker introduces a scenario where a user is trying to understand the codebase in Claude Code by asking about the function of "browser-use."
  • Before answering, Claude reads through the project README and multiple files to build a mental model, which incurs costs for each read.

Limitations of Claude Sessions

  • Each new session starts from scratch; Claude has no memory of previous sessions, leading to repeated reading and searching.
  • This process is likened to hiring a new employee who must learn everything anew with every session.

Introduction of Graphify

  • Graphify is introduced as a tool that builds a knowledge graph after reading the project once, allowing Claude to reference this graph in future sessions instead of re-reading files.
  • The speaker emphasizes that Graphify can be used not only for code but also for various types of documents like research papers and meeting recordings.

How Graphify Works

  • The first pass involves analyzing code structure using a parser that reads all relevant syntax without external API calls.
  • The second pass transcribes audio/video content locally using Faster-Whisper, ensuring privacy and efficiency.
  • The third pass scans other document types (markdown, PDFs), extracting concepts and relationships while merging everything into one cohesive graph.

Efficiency Gains with Graphify

  • A comparison between two identical sessions shows that using Graphify reduces token usage from 120,000 to 113,000 tokens—an approximate 8% savings.
  • Despite minor token savings, the quality of responses improves significantly when using Graphify; detailed explanations are provided compared to simpler mentions without it.

Setting Up Graphify

  • Instructions on how to install Graphify are given: users should run specific commands in their terminal based on their operating system.
  • Users are advised on potential errors during installation on Windows due to command syntax differences.

This structured overview captures key insights from the transcript while providing timestamps for easy navigation.

Graphify: Enhancing Project Management with Graphs

Overview of Graphify's Functionality

  • The process involves analyzing 357 code files, 53 documents, and seven images, requiring selection of subdirectories due to a threshold limit of 200 files.
  • After approximately 12 minutes, the output includes a directory named graphify-out, which contains a GRAPH_REPORT.md file detailing the analysis results.
  • The report identifies 4,041 nodes (concepts), 20,900 edges (relationships), and 185 communities within the project structure.

Understanding the Graph Visualization

  • The graph visualization displays communities on the right side; users can select or deselect these communities for focused analysis.
  • Each dot represents a concept such as classes or functions, while lines indicate relationships. Dot size correlates with connectivity—larger dots signify more connections.

Benefits of Using Graphify

  • Maintaining an updated graph is crucial as projects evolve; users can run graphify update to refresh changes or set up git hooks for automatic updates upon commits or branch switches.
  • When starting a new session in Claude (the AI tool), it automatically reads the graph report without needing additional commands from users.

Efficiency Metrics and Real-world Application

  • A benchmark comparison shows that loading all files directly into Claude costs around 123,000 tokens versus only 1,700 tokens when using Graphify’s method—yielding a savings ratio of about 71.5 times.
  • However, this benchmark may not reflect typical user behavior since most do not load large numbers of files at once; practical savings are more significant in mixed projects or lengthy sessions.

Broader Applications Beyond Code Projects

  • Graphify proves beneficial beyond coding contexts; it was successfully applied to YouTube scripts and research notes without programming languages involved.
  • Users can leverage Graphify for various document types like PDFs and meeting recordings to create interconnected maps that enhance information retrieval across diverse content types.
Video description

Claude Code / Antigravity + Graphify = Instant Knowledge Graph https://github.com/safishamsi/graphify For business enquiries: ai@futurminds.com 🔗 My Resources: Use the below link and coupon code 'FUTURMINDS ' to get 10% DISCOUNT. Hostinger VPS : https://www.hostg.xyz/SHJ4D Start creating n8n workflows: https://n8n.partnerlinks.io/futurminds ----------- Claude Code rebuilds your entire project understanding from scratch every session — re-reading the same files, burning tokens, and starting blind every time you open a new chat. Graphify fixes this: it maps your project once into a knowledge graph, and Claude reads that graph automatically at the start of every session instead of re-reading your files. In this tutorial, I break down exactly how Graphify works (three passes — zero tokens for code, one-time cost for docs), run a real 10-question test in two identical Claude Code sessions to measure the actual token savings, and walk through the full setup from scratch. I also break down what the "71.5x fewer tokens" benchmark actually measures — and when the savings are real for your project. ⏰ Timestamps: 0:00 Why Claude Code Reads Your Files Every Session 1:14 What Graphify Does for Claude Code 1:59 Graphify Architecture: 3 Passes Explained 3:37 Real Token Savings Test: 10 Questions 5:06 How to Install Graphify Step by Step 6:24 Graph Output and How to Update It 7:59 Real-World Impact + 71x Claim Debunked 9:36 Graphify for Research, Content, and Business Folders 🔑 Key Takeaways: • Run pip install graphifyy — double y (the single-y package on PyPI is a completely different tool) • Pass 1 (code parsing) runs entirely on your machine with zero token cost — only Pass 3 touches Claude's API, and only once • Graph overhead means the first 2-3 questions cost slightly more; savings compound after the crossover point • The 71.5x benchmark compares against pasting all files into context at once — a workflow almost nobody uses in real Claude Code sessions ❓ FAQ: Q: Does Graphify actually save tokens in Claude Code? A: Yes, but not 71x in a typical session. In a real 10-question Claude Code session on the browser-use project, Graphify used 113,000 tokens vs 120,000 without — about 7-8% savings. Savings compound with session length, project size, and mixed code+docs projects. Q: How do I install Graphify for Claude Code? A: Run pip install graphifyy (double y — the single-y package is a different tool), then graphify install to register the always-on hook, then /graphify inside your Claude Code session to build the graph. Full walkthrough at 5:06. Q: What is a knowledge graph in the context of Graphify? A: Graphify's knowledge graph is a map of nodes (functions, classes, documents, concepts) and edges (relationships like calls, imports, references), organized into neighborhoods. Claude reads a summary of this graph at the start of every session via a PreToolUse hook — replacing dozens of file reads with a single structured map. Q: Does Graphify work on non-code projects? A: Yes. Graphify works on any folder — research papers, meeting recordings, strategy documents, content vaults. Pass 1 (code parsing) is skipped on text-only projects, but Pass 3 still extracts concepts and builds a navigable graph from markdown, PDFs, and documents. 🔗 Resources & Links: - [LINK: Graphify GitHub Repository — install instructions and documentation] - [LINK: browser-use GitHub — the project used in this demo] - [LINK: FuturMinds Claude Code Tutorials Playlist] 📢 Subscribe to FuturMinds for weekly AI tool tutorials and automation guides — helping non-dev builders master AI systems. Whether you're looking for a complete Graphify setup guide, trying to reduce Claude Code token usage on a growing project, or want to understand how knowledge graphs improve AI coding sessions, this walkthrough covers everything from install to real benchmark analysis.