AI BriefingCodebase Graph
Key point
The Graphify skill for Claude Code turns codebases into a knowledge graph and claims to greatly reduce token usage.
Details
Running /graphify reads codebases, notes, PDFs, screenshots, diagrams, whiteboard photos, images, video, and audio and combines them into a single knowledge graph. Running graphify . can process any folder, and code parsing supports 25 languages including Python, JS/TS, Go, Rust, and Java.
Processing happens in 3 stages.
- Code files are parsed with tree-sitter AST to extract classes, functions, imports, call graphs, docstrings, and rationale comments.
- Video and audio are transcribed locally with faster-whisper, attaching domain-aware prompts drawn from the corpus.
- Claude subagents extract concepts and relationships from documents, images, and transcripts in parallel.
The output is saved as graph.html, GRAPH_REPORT.md, graph.json, and cache/. Relationships are tagged as EXTRACTED, INFERRED, or AMBIGUOUS to separate directly observed information from inference. Clustering is based on Leiden, and no embeddings or vector DB are used.
The author claims it uses 71.5x fewer tokens per query compared to reading raw files directly. Installation involves adding graphifyy via uv, pipx, or pip, then integrating it with each platform using the appropriate graphify install command for Claude Code/Codex/OpenCode/Cursor/Gemini CLI/Copilot CLI, etc.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Whiteboard Launches Open-Source VS Code Fork for Human-AI Collaborative Software Design