coilmaps

A modern codebase is most easily understood the way a city is most easily understood: from above. Coilmaps reads a Python repository and draws its data architecture as something you can actually look at. The point is not the picture; the point is the question the picture provokes. Where the load is, where the cycles are, where the thing you thought was small turned out to be the centre of gravity.

Architecture diagrams are one of the few engineering artefacts whose half-life is measured in days. The team finishes the diagram, ships the next feature, and the diagram quietly becomes a lie. Most engineers stop trusting them around month three.

Coilmaps is the small attempt to fix the asymmetry: a diagram that is rebuilt every time you ask for it, generated directly from the code rather than from anyone’s memory of the code.

Point coilmaps at a public GitHub repository and it parses the source (modules, classes, imports, call graphs) and renders the result as an interactive map. Three.js handles the rendering; the heavy lifting is in the static analysis underneath. An LLM advisor sits alongside the map and answers questions about it in plain language.

The tool is deliberately read-only. It does not refactor, it does not lint, it does not nag. It draws the picture and gets out of the way.

In small repositories the map confirms what the engineer already suspected; in large repositories it surfaces structures the engineer did not know existed. Both outcomes have proven useful: the first as a tool for new joiners, the second as a tool for the senior engineers who have stopped reading the codebase top to bottom because there is no longer a top.

Coilmaps is live at coilmaps.com against public repositories. Private-repository support, language coverage beyond Python, and a small set of architectural-anti-pattern detectors are next.