SchemaGen

SchemaGen is a TypeScript CLI that turns schema source files into a diagram using OpenAI and PlantUML.

It is aimed at database-schema-heavy codebases. The CLI scans a target directory, finds the files that actually define the schema, generates valid PlantUML, and then either:

• saves a rendered svg or png
• opens the generated diagram in the PlantUML web editor

Installation

Install the package via npm:

npm install erd-schema-gen

Run the generator using:

node node_modules/.bin/schemagen

Features

• Automatically discovers schema-defining files instead of requiring a manual file list
• Works well with the Prisma and Drizzle examples included in this repo
• Generates PlantUML using the local docs in docs/ as a syntax reference
• Supports svg and png output
• Can open either the rendered file or the PlantUML editor automatically

Requirements

• Node.js 18 or newer
• An OpenAI API key
• Internet access to OpenAI
• Internet access to www.plantuml.com for rendered image output
• A desktop environment if you want the CLI to auto-open the result

Configuration

SchemaGen reads the API key from either:

• --open-ai-key <string>
• OPEN_AI_KEY in a local .env file

Example:

OPEN_AI_KEY=sk-...

Options

| Option | Description | | --- | --- | | --open-ai-key <string> | OpenAI API key. If omitted, SchemaGen reads OPEN_AI_KEY from .env. | | --dir <string> | Directory to scan. Defaults to the current working directory. | | --output <string> | Output directory or exact file path. Defaults to the current working directory. | | --type <string> | Output format: svg or png. Defaults to svg. | | --editor | Open the generated PlantUML in the PlantUML web editor instead of saving a local file. |

Output behavior

• If --output points to a directory, SchemaGen writes ERD_diagram.<type> into that directory.
• If --output ends with .svg or .png, SchemaGen writes to that exact file path.
• Use relative paths for --dir and --output. The current implementation resolves both relative to the current working directory.

Examples

Generate an SVG from the Prisma example:

node node_modules/.bin/schemagen --dir examples/prisma

Generate a PNG into a custom directory:

node node_modules/.bin/schemagen --dir examples/drizzle-orm --output diagrams --type png

Write to an explicit file path:

node node_modules/.bin/schemagen --dir examples/drizzle-orm --output diagrams/schema.svg

Open the PlantUML editor instead of downloading a file:

node node_modules/.bin/schemagen --dir examples/drizzle-orm --editor

How It Works

1. Builds a simplified tree of the target directory and skips noisy folders such as node_modules, .git, .idea, .vscode, dist, build, out, and logs.
2. Uses gpt-5.4-nano to identify which files directly define the schema.
3. Reads those files and asks gpt-5.4-nano to produce valid PlantUML.
4. Uses the markdown docs in docs/ to verify PlantUML syntax before generating the diagram.
5. Either opens the PlantUML editor or renders the final diagram through PlantUML and saves it locally.

What It Works Well With

• Prisma schema files such as schema.prisma
• Drizzle table definitions and relation files
• Other ORM or entity files that directly declare persisted models and relationships

Limitations

• File discovery is AI-assisted, so accuracy depends on how clearly the schema is expressed in the codebase.
• Rendering a local image requires network access to PlantUML.
• The CLI opens the editor or the generated file automatically.
• The primary use case is schema and ERD generation. The PlantUML generation step can choose a class diagram when the input fits better, but the tool is optimized around database structure.

Development

Useful commands:

pnpm dev
pnpm build
pnpm start -- --dir examples/prisma

Notes:

• pnpm dev compiles the project and runs the CLI against examples/drizzle-orm with --editor.
• The PlantUML reference material used by the generator lives in docs/.

Publish to npm

npm login
npm publish --access public

License

MIT