Structura

Problem

Diagramming tools tend to fall into one of two camps. Text-first tools like PlantUML and Mermaid are fast to write but demand you hold the whole syntax in your head. Canvas tools are immediate but turn structure into pixel positions — the diagram stops being data, and editing it later means dragging boxes around again.

Neither is a good fit for "I know the entities and the relationships, I want a correct UML or ER diagram out of it".

Key technical decision, and the trade-off

Decision: model the diagram as structured data first, and treat rendering as a pure function of that data. The UI is a form editor over the model, not a canvas. Users declare entities, attributes and relationships in a shaped interface, and the diagram is generated from the resulting document.

Why: structure survives round-tripping. Because the source of truth is a typed document rather than a set of coordinates, the same model can be re-rendered at a different layout without the user redoing work, and it can be exported to PlantUML for anyone who wants the text form.

The trade-off: freedom of layout. A structured editor cannot place a box in an arbitrary position, so genuinely irregular diagrams — the ones where position carries meaning — are outside its scope. I took the constraint deliberately: the class of diagram the tool targets is regular by nature, and in exchange every diagram is valid by construction.

Architecture

Express app

incoming request

→

rateLimiter(opts)

middleware, one line

→

Sliding window

prune → count → decide

Result

Has predefined layouts

What I'd do next

• Add a keyboard-first editing path with command shortcuts for creating entities and relations.
• Persist work in progress locally so a refresh never loses a diagram.
• Add SVG and PNG export with a chosen theme.
• [PLACEHOLDER: what you would actually build next on Structura]