graphify: a knowledge graph generator for codebases and documentation that integrates with AI coding assistants

What it solves

Graphify transforms a codebase and its associated documentation (PDFs, images, videos) into a searchable knowledge graph. This allows developers to query the relationships between concepts, trace paths between different parts of the code, and understand the architecture without manually grepping through files or relying solely on vector-based RAG.

How it works

Graphify uses a hybrid approach to map projects:

• Code Parsing: Uses tree-sitter AST for deterministic, local parsing of code across approximately 40 languages. This process is entirely local and does not require an LLM.
• Semantic Pass: For non-code assets like PDFs, images, and videos, it uses an LLM (via a configured API key or the AI assistant's model) to perform a semantic analysis.
• Graph Generation: It produces a graph.json file containing the full graph, a graph.html for interactive visualization, and a GRAPH_REPORT.md highlighting key concepts and connections.
• AI Assistant Integration: It can be registered as a "skill" for various AI coding assistants (like Claude Code, Cursor, and GitHub Copilot), enabling the assistant to query the graph directly using commands like /graphify.

Who it’s for

Software engineers and architects who need to navigate and understand complex codebases, especially when onboarding to a new project or performing deep architectural analysis.

Highlights

• Local-first code mapping: Code is parsed locally via tree-sitter, ensuring privacy and no LLM costs for code analysis.
• Graph-based navigation: Unlike vector indices, it provides a real graph that can be traversed to find shortest paths between concepts.
• Multi-modal support: Maps code, docs, PDFs, images, and video/audio into a single unified knowledge graph.
• Broad assistant compatibility: Integrates with over 20 AI coding assistants through custom skills and hooks.
• Confidence tagging: Every edge in the graph is tagged as either EXTRACTED (explicit in source) or INFERRED (derived by resolution).
• Citations: Extracts design rationale from # NOTE:, # WHY:, and # HACK: comments as first-class nodes.