Knowledge Graphs for AI Coding Assistants
Graphify turns code, docs, PDFs, screenshots, diagrams, and transcripts into one queryable graph so your assistant can answer architecture questions with structure instead of guesswork.
pip install graphifyy && graphify installMixed corpus benchmark from the official project examples.
Tree-sitter parsing across modern programming languages.
graph.html, graph.json, GRAPH_REPORT.md.
Open-source, commercial-friendly, telemetry-free by default.
Architecture at a glance
Imports, calls, classes, docstrings, and rationale comments stay attached to the same graph.
Query the graph weeks later without rereading the entire repository from scratch.
Share a human-readable report with teammates or load the graph into your own tooling.
Built for the messy reality of modern codebases
Keyword search can locate files. Graphify explains how those files, decisions, and artifacts connect.
Multimodal extraction
Read code, docs, PDFs, screenshots, diagrams, transcripts, and rationale in one pass.
Structure-first graph build
Combine AST edges, semantic links, and design notes into a queryable NetworkX graph.
Community clustering
Leiden clustering groups subsystems by topology, not by another embedding layer.
God nodes and surprises
Surface architectural gravity wells and unexpected cross-file connections worth investigating.
Assistant-native commands
Designed for Claude Code, Codex, OpenCode, Cursor, Gemini CLI, and other terminal assistants.
Secure-by-design
Strict input validation, bounded downloads, path containment, and escaped output labels.
Query the why, not just the where
Graphify uses static analysis for hard structure, then enriches the corpus with semantic edges, rationale, and multimodal context. Instead of pushing every question through a fresh vector retrieval loop, it gives your assistant a durable graph of the system.
That means path explanations, community views, god nodes, and surprise edges are first-class outputs, not accidental artifacts of one lucky prompt.
Detect
Collect code, prose, visual references, and media into one corpus.
Extract
Use Tree-sitter and semantic extraction to produce nodes, edges, and rationale.
Build
Merge everything into a persistent graph instead of a disposable prompt context.
Cluster
Reveal subsystems with Leiden communities and highlight graph centers.
Explain
Answer graph queries, path lookups, and architectural why-questions.
Export
Ship graph.html, graph.json, GRAPH_REPORT.md, and incremental cache artifacts.
Graph intelligence, not another pile of chunks
One command to start graphing a repository
Use the official package name on PyPI, then install the assistant integration and point Graphify at any folder of code, notes, papers, or diagrams.
pip install graphifyy && graphify install/graphify .Straight answers for technical buyers
What do you install?+
The official PyPI package is graphifyy, while the command you run stays graphify. After install, you can call it from your coding assistant.
Does Graphify send raw source code to a third-party model?+
The official project states that Graphify only sends semantic descriptions to the model already configured in your assistant, not raw source files.
Which assistants work well with it?+
Official materials list Claude Code, OpenAI Codex, OpenCode, Cursor, Gemini CLI, OpenClaw, and several terminal-native assistants.
What artifacts do I get back?+
A visual graph, a durable graph.json, a human-readable report, and a cache that makes repeat runs cheaper.