Using Graphify to Map Codebases for Claude Code

Chase AIgo watch the original →

Graphify creates a deterministic knowledge graph of codebases to provide AI coding agents with a structured map, reducing the need for expensive file-crawling and lowering token usage.

The Breakthrough

Graphify replaces standard grep-based file searching in AI coding agents with a deterministic knowledge graph that maps code structure, dependencies, and documentation, allowing agents to query the repository architecture directly rather than scanning raw files.

What Actually Worked

  • Deterministic Parsing: The tool uses Tree-sitter to parse code locally without an LLM, extracting classes, functions, imports, and call graphs to build the initial graph structure.
  • Multi-Pass Extraction: The system executes three distinct passes: a structural code analysis (pass one), audio/video transcription via Faster-Whisper (pass two), and semantic analysis of documentation and images (pass three).
  • Persistent Hooking: Running graphify hook install enables automatic graph rebuilding after each git commit, ensuring the knowledge map stays synchronized with repository changes without additional API costs.
  • Agent Integration: The tool installs a custom skill for Claude Code, allowing users to trigger queries via /graphify or set the agent to use the graph automatically for all context retrieval.

Before / After

  • Token Usage: In a test tracing a design request flow through the Open Design repository, the standard grep-based approach consumed approximately 200,000 tokens, while the Graphify-enabled approach consumed 80,000 tokens, representing a 60% reduction in token cost.

Context

AI coding assistants often struggle with large repositories because they rely on searching through individual files, which is inefficient and token-heavy. Graphify acts as a middle ground between simple note-taking tools like Obsidian and complex RAG infrastructure. By building a persistent map of nodes and edges, it provides the agent with a structural understanding of the codebase, which remains useful across multiple sessions and team-based workflows.

Content References

{"type": "tool", "title": "Graphify", "url": "https://github.com/safishamsi/graphify", "context": "recommended"}, {"type": "tool", "title": "Claude Code", "context": "mentioned"}, {"type": "tool", "title": "Obsidian", "context": "mentioned"}, {"type": "tool", "title": "Faster-Whisper", "context": "mentioned"}

  • #ai
  • #dev-tooling
  • #claudecode

summary by google/gemini-3.1-flash-lite. probably wrong about something. check the source.