Project Overview

The problem. When you ask Cursor or Claude to refactor a function, it reads your files. It doesn’t understand your codebase. It guesses relationships, misses queue consumers, overlooks DB writes  and confidently suggests changes that silently break things elsewhere. The root cause isn’t the model. It’s that AI reads files. It needs to understand graphs.

What I built. Nomik is an open-source CLI that scans any codebase once and builds a persistent knowledge graph in Neo4j  every function, route, database operation, cron job, and import becomes a node or edge. Your AI queries that graph directly through 21 MCP tools instead of grepping raw files.

The result. Ask “what breaks if I rename processPayment?” and instead of a guess you get: 5 callers, 2 DB tables, 1 cron job — with exact file paths and line numbers. Ask “where is auth handled?” and get the full chain from route → middleware → session store.

Why it matters. On large TypeScript monorepos I kept hitting the same wall: the AI would confidently suggest refactors that broke things it didn’t know existed — a queue consumer calling the same function, a retry job hitting the same DB write. Context delivery was the bottleneck. A knowledge graph fixes that permanently.

Technical scope. Nomik supports 7 languages, extracts 37 node types, and exposes 33 CLI commands. It runs entirely locally — no code leaves your machine. It connects to Cursor, Windsurf, Claude Desktop, and Antigravity with a single command: nomik setup-cursor.