Turn your repo into a knowledge graph your AI assistant can query

What it does

Graphify scans a directory of code, SQL schemas, documentation, PDFs, images, or videos and compiles them into a structured knowledge graph. It then hooks into popular AI coding assistants—Claude Code, Cursor, Codex, Gemini CLI, and roughly fifteen others—so the assistant queries relationships in the graph instead of reading files linearly. The output is a trio of artifacts: an interactive graph.html, a GRAPH_REPORT.md summary, and a machine-readable graph.json.

The interesting bit

The project acts less like a standalone app and more like a universal peripheral for AI assistants, installing platform-specific skills that teach the assistant to prefer graph queries over file grepping. On Claude Code and Gemini CLI it even intercepts search hooks to nudge the model toward the graph automatically, and it can export Mermaid call-flow diagrams for human architecture review.

Key highlights

• Multi-modal ingestion: handles code, shell scripts, R, SQL schemas, PDFs, Office docs, images, and video/audio transcripts
• Broad assistant support: Claude Code, Codex, OpenCode, Cursor, Gemini CLI, GitHub Copilot, Aider, and others
• Dual-use outputs: an interactive graph.html for browsing, a GRAPH_REPORT.md with “god nodes” and surprising connections, and graph.json for the assistant
• Optional backend extras: Neo4j export, Leiden community detection, Ollama local inference, and multiple cloud LLM providers
• Project-scoped installation: can commit assistant skill files directly into the repo for shared team behavior

Caveats

• The PyPI package is named graphifyy (double-y); the README explicitly warns that other graphify* packages are unaffiliated
• Leiden community detection works only on Python < 3.13
• The BYOND DreamMaker (dm) extra may require a C compiler and python3-dev if no prebuilt wheel exists for your platform

Verdict

Worth a look if you manage a large or polyglot codebase and want your AI assistant to stop treating it like a flat text dump. Skip it if you already have a bespoke RAG pipeline you trust, or if your project is small enough that find and grep still feel instant.