Introduction

Schematex renders every diagram a doctor, engineer, or lawyer would actually use — clinical genograms, IEC 61131-3 ladder logic, NSGC pedigrees, IEEE 315 single-line diagrams, cap tables, and many more. A tiny text DSL in; standards-compliant SVG out.

Free. Fully open source. Made for AI. AGPL-3.0, zero runtime dependencies, and a DSL designed around how LLMs actually write text — paste output from ChatGPT or Claude and get a professional diagram back on the first try.

What can I draw?

Schematex draws 45 diagram types across 14 domains — each built to a published standard:

Quick example

Use with AI

Schematex ships a tool layer built for LLMs. Connect Claude.ai to the hosted MCP server in 10 seconds and let it generate validated diagrams in any conversation — no install, no setup.

Or build your own AI feature on top of the Vercel AI SDK. See Use Schematex with AI.

Next steps

• Getting started — install, import, render
• Use Schematex with AI — MCP + Vercel AI SDK + agent loop
• Playground — try any diagram type live
• API reference

How it works

Each diagram type follows the same pipeline:

Text DSL ──→ Parser ──→ AST ──→ Layout Engine ──→ LayoutResult ──→ SVG Renderer ──→ SVG string

Every diagram implements a DiagramPlugin with four methods:

interface DiagramPlugin {
  detect(text: string): boolean;           // auto-detects diagram type
  parse(text: string): DiagramAST;         // text → typed AST
  layout(ast, config): LayoutResult;       // AST → positioned nodes/edges
  render(layout, config): string;          // LayoutResult → SVG string
}

Layout algorithms are domain-specific by design — genogram uses a generation-based layered layout, ecomap uses radial/polar, logic gates use DAG topological sort, ladder logic uses a fixed power-rail layout. Generic layout engines (dagre, ELK) cannot produce standards-compliant output for these diagram types.

Design principles

1. Diagrams professionals actually use — each diagram type implements a published domain specification: McGoldrick 2020 (genogram), Hartman 1978 (ecomap), Bennett 2022 (pedigree), Moreno 1934 (sociogram), IEEE Std 91 (logic gates), IEC 61131-3 (ladder logic), IEEE 315 (SLD), and more.
2. Free & fully open source — AGPL-3.0, zero runtime dependencies (no D3, no dagre, no parser generators), everything hand-written. Small bundle, no supply-chain risk. Commercial license available for closed-source use.
3. Made for AI — the DSL is designed around how LLMs actually write text (CJK quotes, nesting ambiguity, AI-readable errors). Paste output from ChatGPT or Claude and get a professional diagram back on the first try.
4. Semantic SVG output — every SVG includes <title> and <desc> for accessibility, CSS classes for theming, and data-* attributes for interactivity.

Theming

Three built-in presets: default (blue-gray), monochrome (black-and-white / print), dark (Catppuccin Mocha).

Override any token via CSS custom properties injected into each SVG's <style> block:

--schematex-stroke: #1a1a1a;
--schematex-fill: #f5f5f5;