qx

A teeny tiny, type-safe and dependency-free ORM for TypeScript and JavaScript.

Built for you who wants a simple, small ORM that just works.

import * as sqlite from '@rwillians/qx/bun-sqlite';
import { create, from, into, table } from '@rwillians/qx';

const users = table('users', t => ({
  id: t.integer().autoincrement().primaryKey(),
  name: t.string(),
  email: t.string(),
  createdAt: t.datetime(),
}));

// ...

const db = sqlite.connect('./db.sqlite');

await create.table(users).onto(db);

// ...

const rows = await into(users)
  .insert({ name: 'John Doe', email: '[email protected]' })
  .insert([ userA, userB, userC ])
  .run(db);

// ...

const user = await from(users.as('u'))
  .where(({ u }) => e.eq(u.id, 1))
  .one(conn);

if (!user) {
  throw new Error('User not found');
}

console.log(user);

• See roadmap to v1.0.0.
• See more examples.

Vision

Here's the basics of what I need from an ORM, thus it's my priority to build it first:

• Defining model fields should be very similar to defining a schema with Zod (with support for validation refiements and transformations).
• The model should have a schema compatible with standard schema, meaning it should be interoperable with Zod, ArkType, Joi, etc.
• It should have a SQL-Like, type-safe, fluent query builder api that works even for NoSQL databases¹, allowing us to write queries once then use them with any supported database.
• The query builder should output a plain map representation of the query that can be encoded to JSON, mostly for three reasons:
  1. It's easy to test;
  2. Makes it easier to debug queries; and
  3. Makes qx more modular, allowing the community to build their own extensions.
• The query results should be type-safe.

¹ Some database adapters might not support all query features, that's expected.

Once this vision is fullfilled, qx will become v1.0.0.

Components

The vision above implies the existence of four main components to this library:

1. A table factory that outputs a model with a standard schema;
2. A query builder that outputs a plain map representation of the query;
3. A query engine that orchestrates the query execution using a database adapter; and
4. Database adapters that can execute queries for a specific database driver.

Database Adapters

Database adapters are per driver implementation. Quex ships with a few hand picked built-in database adapters:

• [x] bun-sqlite3 (prioritary)
• [ ] bun-postgres
• [ ] mongodb

For community-built adapters, check GitHub's tag #qx-adapter (you won't find anything there yet).

Examples

Here are some examples that I'm using to guide the implementation.

Define a table:

// src/db/tables/backups.ts
import { z } from 'zod/v4';
import { type as arktype } from 'arktype';
import { defineColumn, create, table } from 'qx';

// custom types
const tc = {
  absolutePath: () => defineColumn({
    type: 'VARCHAR',
    schema: z
      .string()
      .refine(str => str.startsWith('/'), "must be an absolute path")
      .transform(str => str.endsWith('/') ? str.slice(0, -1) : str)
  }),
  bytes: () => defineColumn({
    type: 'INTEGER',
    schema: arktype('number.integer > 0'),
  }),
  email: () => defineColumn({
    type: 'VARCHAR',
    schema: z.string().email(),
  }),
};

export const backups = table('backups', t => ({
  id: t.integer().autoincrement().primaryKey(),
  parentId: t.integer().nullable(),
  state: t.enum(['succeeded', 'failed']).default('succeeded'),
  path: tc.absolutePath(),
  size: tc.bytes().nullable(),
  notifyableContacts: tc.email().array().default([]),
  //                            ↑ should be stored as VARCHAR[] in postgres
  //                              should be stored as json encoded TEXT in sqlite
  createdAt: t.datetime().default(() => new Date),
}));

export type Backup = typeof backups.infer;
export type BackupForInsert = typeof backups.inferForInsert;
export type BackupForUpdate = typeof backups.inferForUpdate;

Create the table in the database:

import * as sqlite from 'qx/bun-sqlite';

// ...

const db = sqlite.connect('./db.sqlite');

await create.table(backups).onto(db);

Insert rows:

import { into } from 'qx';
import * as sqlite from 'qx/bun-sqlite';

// ...

const db = sqlite.connect('./db.sqlite');

const rows = await into(backups)
  .insert({ state: 'succeeded', path: '/data/backup_1.tar.gz', size: 104857600 })
  .run(db);

Query the table:

import { expr, from } from 'qx';
import * as sqlite from 'qx/bun-sqlite';

const conn = sqlite.connect('./db.sqlite');

// ...

const yesterday = new Date(Date.now() - (24 * 60 * 60 * 1000));

const results = await from(backups.as('b1'))
  .leftJoin(backups.as('b2'), ({ b1, b2 }) => expr.eq(b2.id, b1.parentId))
  .where(({ b1, b2 }) => expr.and([
    expr.eq(b1.state, 'failed'),
    expr.gte(b1.failedAt, yesterday),
    expr.eq(b1.scheduledBy, '[email protected]'),
  ]))
  .orderBy(({ b1 }) => expr.desc(b1.scheduledAt))
  .limit(25)
  .offset(0)
  .select(({ b1, b2 }) => ({
    ...b1,
    parentPath: b2.path,
    totalSizeMiB: expr.div(expr.add(b1.size, expr.coalesce(b2.size, 0)), 1048576),
  }))
  .all(db);

No singleton magic here! Not on my watch. You need to explicitly pass the db connection to the query.

The results would look like this:

[
  {
    id: 2,
    parentId: 1,
    state: 'failed',
    path: '/backups/20251130133100.tar.gz',
    size: 104857600,
    notifyableContacts: ['[email protected]'],
    createdAt: new Date('2025-11-30T13:31:00.000Z'),
    parentPath: '/backups/20251030134200.tar.gz',
    totalSizeMiB: 42069,
  }
]