Graphify Hits 450K Downloads and 40K Stars in 26 Days

Graphify is an open-source tool that builds a persistent knowledge graph for AI coding assistants. It reads codebases, documentation, PDFs, images, and even audio/video to map out relationships as a graph, leaving behind a browser-ready graph.html, a summary report GRAPH_REPORT.md, a reusable graph.json, and a SHA256 cache/.

• It reached 450K downloads and 40K stars in just 26 days.
• The first stage extracts classes, functions, imports, call graphs, docstrings, and rationale comments from code files using AST, with no LLM involved.
• The second stage transcribes audio/video locally with faster-whisper, and the transcripts are cached.
• The third stage has Claude subagents extract concepts and relationships in parallel from documents, images, and transcripts.
• The graph is merged using NetworkX and clustered with Leiden community detection, without any separate embedding step or vector DB.
• Relationships are labeled EXTRACTED, INFERRED, or AMBIGUOUS to distinguish confirmed facts.
• For SQL, tables, views, functions, foreign keys, and FROM/JOIN relationships are extracted using a deterministic AST.
• .graphifyignore follows .gitignore syntax and ensures scanning doesn't cross VCS boundaries.
• It integrates with multiple agents/IDEs including Claude Code, Codex, Cursor, Gemini CLI, and Copilot CLI.