kysely-typegen

Generate Kysely type definitions from your database.

Thank you kysely-codegen for inspiration and ideas!

Why?

Why kysely-typegen if there is already kysely-codegen? Comparison:

| | [email protected] | kysely-typegen | | ----------------------------- | ---------------------------------------- | ----------------------------------------------------------------------------------- | | Install Size | 6.8 MB | 22.1 kB | | Dependencies | 35 total | 0 (no runtime dependencies) | | Type | CLI | Library/Programmatic Usage | | Code Size/Maintainability | Heavy | Lightweight/Simple and straightforward (string manipulation instead of complex AST) | | Database Support | PostgreSQL, MySQL, SQLite, MSSQL, LibSQL | PostgreSQL, MySQL, SQLite (can be easily extended to more) |

kysely-typegen is a library (not a CLI), which means you are in control of where and how to run it, and is designed to be extensible, easy to add support for more database dialects.

For example: you can use Node.js with the --env-file CLI option, dotenv dependency is not required (but can be used), you are in control.

Note: kysely-typegen doesn't have the same features and customization as kysely-codegen, it has less features to keep it simple and lightweight, but can be extended to your own needs. For more details, why this project was created, and the design decisions, see: https://github.com/RobinBlomberg/kysely-codegen/issues/175.

Prerequisites

Node.js >= 24.0.0

Installation

npm install --save-dev kysely-typegen

Peer dependencies:

npm install kysely

Plus a Kysely dialect driver for your database (see Setup Kysely database below).

Usage

Setup Kysely database

Create your Kysely database instance and export a databaseTypegen for the script in the next section. The rest of the guide is dialect-agnostic: only this file changes per database.

PostgreSQL

KyselyTypegenPostgresDialect works with any Kysely PostgreSQL dialect, including the built-in PostgresDialect (using pg) and PostgresJSDialect (using postgres). The example below uses kysely-postgres-js.

npm install kysely-postgres-js postgres

// database.ts
import { Kysely } from "kysely"
import { PostgresJSDialect } from "kysely-postgres-js"
import { KyselyTypegenPostgresDialect } from "kysely-typegen/postgres"
import postgres from "postgres"

import type { DB } from "./codegen.ts"

const dialect = new PostgresJSDialect({
  postgres: postgres({
    database: process.env["DATABASE_NAME"] ?? "database",
    host: process.env["DATABASE_HOST"] ?? "localhost",
    user: process.env["DATABASE_USER"] ?? "user",
    password: process.env["DATABASE_PASSWORD"] ?? "password",
    port: Number.parseInt(process.env["DATABASE_PORT"] ?? "5432", 10),
  }),
})

export const database = new Kysely<DB>({ dialect })
export const databaseTypegen = new KyselyTypegenPostgresDialect({ database })

MySQL

npm install mysql2

// database.ts
import { Kysely, MysqlDialect } from "kysely"
import { KyselyTypegenMySQLDialect } from "kysely-typegen/mysql"
import { createPool } from "mysql2"

import type { DB } from "./codegen.ts"

const dialect = new MysqlDialect({
  pool: createPool({
    database: process.env["DATABASE_NAME"] ?? "database",
    host: process.env["DATABASE_HOST"] ?? "localhost",
    user: process.env["DATABASE_USER"] ?? "user",
    password: process.env["DATABASE_PASSWORD"] ?? "password",
    port: Number.parseInt(process.env["DATABASE_PORT"] ?? "3306", 10),
  }),
})

export const database = new Kysely<DB>({ dialect })
export const databaseTypegen = new KyselyTypegenMySQLDialect({ database })

SQLite

npm install better-sqlite3

// database.ts
import Database from "better-sqlite3"
import { Kysely, SqliteDialect } from "kysely"
import { KyselyTypegenSQLiteDialect } from "kysely-typegen/sqlite"

import type { DB } from "./codegen.ts"

const dialect = new SqliteDialect({
  database: new Database(process.env["DATABASE_PATH"] ?? "database.sqlite"),
})

export const database = new Kysely<DB>({ dialect })
export const databaseTypegen = new KyselyTypegenSQLiteDialect({ database })

Generate the type definitions

Create a script that uses databaseTypegen to introspect your database and write the generated types to a file. This script is the same regardless of the underlying database:

// scripts/typegen.ts
// This is an example, you can be more creative:
// time it takes to generate, number of tables/enums generated, etc.
import fs from "node:fs"
import path from "node:path"

import { database, databaseTypegen } from "./database.ts"

const result = await databaseTypegen.typegen()
const codegenContent = result.lines.join("\n")
const codegenPath = path.join(process.cwd(), "codegen.ts")
await fs.promises.writeFile(codegenPath, codegenContent, "utf-8")
await database.destroy()

Run with Node.js (using --env-file to load environment variables, no dotenv dependency required):

node --env-file=.env scripts/typegen.ts

Using the type definitions

Import DB into new Kysely<DB>, and you're done!

import { database } from "./database.ts"

const rows = await database.selectFrom("User").selectAll().execute()
//    ^ { createdAt: Date; email: string; id: number; ... }[]

Fully type-safe queries derived from your actual database schema.

Extending to other database dialects

kysely-typegen ships with KyselyTypegenPostgresDialect, KyselyTypegenMySQLDialect, and KyselyTypegenSQLiteDialect, but you can add support for any database by extending the abstract KyselyTypegenDialect class.

Only one thing is required:

• scalars: a Record<string, string> mapping the database column types to TypeScript types.

import { KyselyTypegenDialect } from "kysely-typegen"

export class KyselyTypegenMSSQLDialect extends KyselyTypegenDialect {
  public override readonly scalars: Record<string, string> = {
    bigint: "Int8",
    bit: "boolean",
    char: "string",
    datetime: "Timestamp",
    decimal: "Numeric",
    int: "number",
    nvarchar: "string",
    smallint: "number",
    text: "string",
    varbinary: "Buffer",
    varchar: "string",
    // ...
  }
}

If your database supports enums, override the optional introspectEnums() hook, which returns two maps:

• named: enum name → values (emitted as export type Name = "a" | "b").
• inline: ${tableName}.${columnName} → values (emitted inline at the column site, for databases like MySQL where enums are anonymous per-column).

import type { IntrospectedEnums } from "kysely-typegen"

protected override async introspectEnums(): Promise<IntrospectedEnums> {
  // Query your database's information schema...
  return { named: [], inline: new Map() }
}

If your database needs to normalize column type spellings before scalar lookup (e.g. strip VARCHAR(255) → VARCHAR, or uppercase keys), override the optional normalizeDataType(dataType: string): string hook.

The base class handles introspection (via Kysely's database.introspection.getTables()), sorting, and the final type generation. Contributions of new dialects are welcome!

Contributing

Anyone can help to improve the project, submit a Feature Request, a bug report or even correct a simple spelling mistake.

The steps to contribute can be found in the CONTRIBUTING.md file.

License

MIT