// the find
Graphify-Labs/graphify
AI coding assistant skill (Claude Code, Codex, OpenCode, Cursor, Gemini CLI, and more). Turn any folder of code, SQL schemas, R scripts, shell scripts, docs, papers, images, or videos into a queryable knowledge graph. App code + database schema + infrastructure in one graph.
Graphify turns any codebase into a queryable knowledge graph using tree-sitter AST extraction for code (no LLM calls, no cost) and LLM extraction for docs and media. You install it as a 'skill' in your AI coding assistant, then query it with natural language instead of grepping files. Aimed at developers who want their AI assistant to actually understand the codebase structure rather than reading files one at a time.
The zero-token graph build for code is real and meaningful — tree-sitter extracts calls, imports, and inheritance across 36 languages locally, so a 100k-file repo costs nothing to map. Every edge carries a confidence tag (EXTRACTED vs INFERRED vs AMBIGUOUS), which is unusually honest for a tool in this space and makes it possible to trust the output selectively. The git merge driver that union-merges graph.json on conflict is a small detail that shows the team actually thought through team workflows. Leiden community detection running without LLM involvement means the 'subsystems' view is fast and reproducible rather than varying by prompt temperature.
The PyPI package is named 'graphifyy' (double-y) while the CLI command is 'graphify' — this is a genuine footgun that the README spends two troubleshooting paragraphs on and suggests the good name was squatted. Leiden community detection requires Python < 3.13, which will quietly break for anyone on a current Python install who doesn't read the extras table carefully. The benchmarks are selectively favorable: graphify wins recall@10 on LOCOMO by a wide margin but loses QA accuracy to supermemory (45.3% vs 49.7%), and a single 50-sample eval for LongMemEval-S isn't statistically convincing. Supporting 20+ AI assistant platforms means 20+ code paths and skill files to keep synchronized — CHANGELOG.md will tell you how often platform-specific bugs creep in.