I built this after spending hours manually tracing a codebase I didn't write — following imports by hand, guessing at blast radius before touching anything, finding a database table involved three levels deep.
The tool statically parses Python (native AST) and JS/TS (tree-sitter) — no LLM guessing, fully deterministic. It builds an interactive call graph with impact analysis, execution flowcharts, a cyclomatic complexity heatmap, AI-generated docs, and code-to-data lineage that traces from a Python function all the way to the specific database column it reads or writes.
Tested against Apache Superset (1,458 files): found 37 tables, 206 columns, and 462 read/write edges automatically with zero configuration.
Three ways to run it:
VS Code extension — backend ships bundled, no Python or uv required
Docker: docker compose up
Local: uv sync + npm install
Everything runs on your machine. No account, no telemetry, no cloud.
Benchmarks against FastAPI (1,138 files), three.js, and NestJS (1,907 files) are in the repo.
Top comments (0)