Zenstack schema modeling

ZenStack — Schema Modeling (ZModel)

ZModel is ZenStack's schema language — a superset of Prisma's schema. The schema lives at zenstack/schema.zmodel by default. For access policies/validation use zenstack-access-control; for queries use zenstack-querying.

After editing the schema:

zen check      # only validate the schema for syntax/semantic errors (use --schema <file> for a non-default path)
zen generate   # regenerate the database client — also validates the schema as it compiles

Run zen check when you just want to confirm the schema is valid; run zen generate when you want to regenerate the client (the normal step after a schema change).

Strings accept both single and double quotes. Every valid Prisma schema is valid ZModel, so existing Prisma knowledge transfers directly — the sections below cover both the basics and the additions ZModel layers on top.

Datasource

Exactly one datasource block per schema. The ORM runtime doesn't use url (the driver carries the connection); the migration engine does.

datasource db {
    provider = 'postgresql'        // 'postgresql' | 'mysql' | 'sqlite'
    url      = env('DATABASE_URL') // or 'file:./dev.db' for SQLite
}

ZModel has no generator block. Code generation is driven by zen generate; optional behavior is configured with plugin blocks (e.g. plugin policy, plugin prisma).

Models & fields

model User {
    id        Int      @id @default(autoincrement())
    email     String   @unique
    name      String?              // optional
    role      Role     @default(USER)
    tags      String[]             // list
    createdAt DateTime @default(now())
    updatedAt DateTime @updatedAt
}

Scalar types: String, Boolean, Int, BigInt, Float, Decimal, DateTime, Json, Bytes, Unsupported("...").

Modifiers: ? optional, [] list. A field cannot be both optional and a list.

Field attributes (@)

@id, @unique, @default(...), @updatedAt, @map('col'), @relation(...), @db.* (native type, e.g. @db.VarChar(64)), @json (typed-JSON field), @computed (virtual field).

Model attributes (@@)

@@id([a, b]) (composite PK), @@unique([a, b]), @@index([a, b]), @@map('table'), @@delegate(field) (polymorphism), plus @@allow / @@deny / @@validate (see zenstack-access-control).

model City {
    country String
    name    String
    @@id([country, name])     // composite primary key
    @@map('cities')
}

A model needs an identifier: a @id field, an @@id([...]), or — failing those — a @unique field / @@unique([...]).

Default value functions

autoincrement(), now(), cuid(), uuid() / uuid(4), ulid(), nanoid(), dbgenerated("..."), plus literals and enum values.

Since v3.1.0, string ID generators accept a format with %s:

id String @id @default(uuid(4, "user_%s"))   // -> "user_<uuid>"

Enums

enum Role {
    USER
    ADMIN
}

Enum value names are global and must not collide.

Relations

The owner side holds the foreign key (@relation(fields: [...], references: [...])).

One-to-one — FK is @unique; the non-owner side is optional:

model User {
    id      Int      @id
    profile Profile?
}
model Profile {
    id     Int  @id
    user   User @relation(fields: [userId], references: [id])
    userId Int  @unique
}

One-to-many — list on the non-owner side:

model User {
    id    Int    @id
    posts Post[]
}
model Post {
    id       Int  @id
    author   User @relation(fields: [authorId], references: [id])
    authorId Int
}

Many-to-many (implicit) — list on both sides; the migration engine creates the join table:

model User { id Int @id  posts Post[] }
model Post { id Int @id  editors User[] }

Many-to-many (explicit) — model the join table yourself when it needs extra fields:

model UserPost {
    user   User @relation(fields: [userId], references: [id])
    userId Int
    post   Post @relation(fields: [postId], references: [id])
    postId Int
    @@id([userId, postId])
}

Named relations disambiguate multiple relations between the same two models (give both ends the same name): @relation('UserPosts', ...).

Self-relations use a named relation, e.g. a one-to-many manager/reports tree:

model Employee {
    id           Int        @id
    managerId    Int?
    manager      Employee?  @relation('Mgmt', fields: [managerId], references: [id])
    subordinates Employee[] @relation('Mgmt')
}

Referential actions on the owner side: onDelete / onUpdate with Cascade, Restrict, NoAction, SetNull, SetDefault.

author User @relation(fields: [authorId], references: [id], onDelete: Cascade)

Custom types & mixins

A type declaration is not backed by a table. Use it for typed JSON (below) or as a mixin of reusable fields, applied with with:

type BaseFields {
    id        String   @id @default(cuid())
    createdAt DateTime @default(now())
    updatedAt DateTime @updatedAt
}

model User with BaseFields {
    email String @unique
}
model Post with BaseFields {
    title String
}

Mixin fields are inlined into the model. Multiple mixins are allowed if names don't conflict. This is ZModel's mechanism for sharing fields across models (use it instead of an abstract base model).

A type may even carry relation fields (v3.6.0+), but then it can only be used as a mixin, not as a JSON field:

type AuditMixin {
    createdBy   User   @relation('CreatedBy', fields: [createdById], references: [id])
    createdById String
}
model Post with AuditMixin { title String }

Polymorphism — @@delegate

Multi-table inheritance: a base model with a discriminator field and @@delegate, concrete models using extends. Concrete and base rows share the same id; you create concrete models, never the base directly.

model Content {
    id        Int      @id @default(autoincrement())
    createdAt DateTime @default(now())
    owner     User     @relation(fields: [ownerId], references: [id])
    ownerId   Int
    type      String   // discriminator
    @@delegate(type)
}

model Post extends Content {
    content String
}
model Image extends Content {
    data Bytes
}

Customize the stored discriminator value with @@delegateMap (v3.7.1+), which also works with enum discriminators:

model Video extends Content {
    url String
    @@delegateMap("video")
}

Strongly-typed JSON

Define a type, use it as a field type, and mark the field @json. The column is plain JSON in the database, but ZenStack validates writes and returns a typed value.

type Address {
    street  String
    city    String
    zip     Int
}
model User {
    id      Int     @id
    address Address @json
}

Computed fields

Declare a virtual field with @computed; implement it with a Kysely expression in the client config. It can be selected, filtered, sorted, and referenced in policies.

model User {
    id        Int @id
    postCount Int @computed
}

const db = new ZenStackClient(schema, {
    dialect,
    computedFields: {
        User: {
            postCount: (eb) =>
                eb.selectFrom('Post')
                    .whereRef('Post.authorId', '=', 'id')
                    .select(({ fn }) => fn.countAll<number>().as('count')),
        },
    },
});

ZModel additions over Prisma

• No generator block (use zen generate + plugins).
• type declarations + with mixins, @@delegate polymorphism, @json typed JSON, and @computed fields are ZenStack additions Prisma lacks.
• Single or double quotes are both fine.

Reference docs

Full ZenStack documentation for this topic is bundled under references/:

• modeling-overview.md — data modeling concepts
• datasource.md — the datasource block
• model.md — models, fields, ids, defaults
• attribute.md — field & model attributes
• relation.md — 1-1, 1-many, many-many, self relations
• enum.md — enums
• custom-type.md — type declarations
• mixin.md — reusable field mixins (with)
• polymorphism.md — @@delegate polymorphic models
• typed-json.md — strongly-typed JSON fields
• computed-fields.md — @computed fields
• zmodel-attribute-reference.md — complete attribute reference
• zmodel-function-reference.md — attribute-function reference