The type-safe data layer for Kysely | Kysera

3 min read Original article ↗

Not an ORM. A data-access toolkit.

SQL you can still see

Kysera is a thin layer over Kysely, not a replacement for it. Every query is still a Kysely query; every escape hatch stays open — drop to the raw instance or a sql template whenever you need to.

One plugin core, two patterns

Plugins intercept queries at the executor, so the Repository pattern and the functional DAL share the same soft-delete filters and RLS policies — in and out of transactions. Repositories add an atomic audit trail on top. Mix both styles in one codebase (CQRS-lite).

Hardened by default

Row-level security enforces SELECT, UPDATE, and DELETE in SQL. Soft delete narrows mutations to live rows. Opt-outs are scoped per statement — there is no global bypass switch to forget about.

Pick a pattern. Or both.

Structured CRUD where you want conventions, composable functions where you want reach — against the same executor, the same plugins, the same transaction.

Repository

users.repository.ts

import {

createORM, createRepositoryFactory, zodAdapter

} from '@kysera/repository'

const orm = await createORM(executor, [])

const users = orm.createRepository(exec =>

createRepositoryFactory(exec).create({

tableName: 'users',

mapRow: row => row,

schemas: { create: zodAdapter(CreateUser) },

})

)

const user = await users.create({

email: 'ada@example.com',

name: 'Ada',

})

await users.softDelete(user.id) // added by the plugin

await users.findAll() // deleted rows filtered

Functional DAL

users.queries.ts

import {

createQuery, createContext, withTransaction,

type DbContext

} from '@kysera/dal'

const userByEmail = createQuery(

(ctx: DbContext<DB>, email: string) =>

ctx.db

.selectFrom('users')

.selectAll()

.where('email', '=', email)

.executeTakeFirst()

)

const ctx = createContext(executor)

await userByEmail(ctx, 'ada@example.com')

// soft-delete filter applied automatically

await withTransaction(executor, async tx => {

await userByEmail(tx, 'ada@example.com')

}) // plugins survive; nested calls → savepoints

Four plugins. No magic.

Each plugin declares a priority tier and runs in a fixed, inspectable order — security → filters → transforms → audit — no matter how you register them.

Batteries included — separately.

Thirteen focused packages. Install what you use; tree-shake the rest.

0

Third-party runtime deps

3

Runtimes: Node, Bun, Deno

Tested in CI against live PostgreSQL and MySQL on every push; concurrency claims are proven by racing tests. A benchmark suite tracks overhead each release: the executor without plugins measures within noise-to-13% of raw Kysely on the execute path, and a full three-plugin stack costs ~15–20%.

Up and running in a minute

Add the packages to an existing project, or let the CLI scaffold one — config, migrations, and a health check included.

Install

npm install kysely zod

npm install @kysera/executor @kysera/repository @kysera/soft-delete

ESM-only, Node 22+. Zod is optional — bring Valibot or TypeBox instead, or skip validation entirely.

Or scaffold a project

npx @kysera/cli init my-app

npx @kysera/cli doctor

init sets up config and migrations; doctor verifies the whole environment in one shot.

Use

import { Kysely, PostgresDialect } from 'kysely'

import { createExecutor } from '@kysera/executor'

import {

createORM, createRepositoryFactory, zodAdapter

} from '@kysera/repository'

import { softDeletePlugin } from '@kysera/soft-delete'

import { z } from 'zod'

const db = new Kysely<Database>({

dialect: new PostgresDialect({ pool })

})

const executor = await createExecutor(db, [softDeletePlugin()])

const orm = await createORM(executor, [])

const users = orm.createRepository(exec =>

createRepositoryFactory(exec).create({

tableName: 'users',

mapRow: row => row,

schemas: {

create: zodAdapter(

z.object({ email: z.string().email(), name: z.string() })

)

},

})

)

const user = await users.create({ email: 'ada@example.com', name: 'Ada' })

await users.softDelete(user.id)

await users.findAll() // soft-deleted rows are filtered out

Keep your SQL. Gain the toolkit.