Use the Json Prisma ORM field type to read, write, and perform basic filtering on JSON types in the underlying database. In the following example, the User model has an optional Json field named extendedPetsData:

model User {
  id               Int     @id @default(autoincrement())
  email            String  @unique
  name             String?
  posts            Post[]
  extendedPetsData Json?
}

Example field value:

{
  "pet1": {
    "petName": "Claudine",
    "petType": "House cat"
  },
  "pet2": {
    "petName": "Sunny",
    "petType": "Gerbil"
  }
}

The Json field supports a few additional types, such as string and boolean. These additional types exist to match the types supported by JSON.parse():

export type JsonValue = string | number | boolean | null | JsonObject | JsonArray;

Reasons to store data as JSON rather than representing data as related models include:

• You need to store data that does not have a consistent structure
• You are importing data from another system and do not want to map that data to Prisma models

You can use the Prisma.JsonArray and Prisma.JsonObject utility classes to work with the contents of a Json field:

const { PrismaClient, Prisma } = require("@prisma/client");

const user = await prisma.user.findFirst({
  where: {
    id: 9,
  },
});

// Example extendedPetsData data:
// [{ name: 'Bob the dog' }, { name: 'Claudine the cat' }]

if (
  user?.extendedPetsData &&
  typeof user?.extendedPetsData === "object" &&
  Array.isArray(user?.extendedPetsData)
) {
  const petsObject = user?.extendedPetsData as Prisma.JsonArray;

  const firstPet = petsObject[0];
}

See also: Advanced example: Update a nested JSON key value

The following example writes a JSON object to the extendedPetsData field:

var json = [{ name: "Bob the dog" }, { name: "Claudine the cat" }] as Prisma.JsonArray;

const createUser = await prisma.user.create({
  data: {
    email: "birgitte@prisma.io",
    extendedPetsData: json,
  },
});

Note: JavaScript objects (for example, { extendedPetsData: "none"}) are automatically converted to JSON.

See also: Advanced example: Update a nested JSON key value

You can filter rows of Json type.

Filter on exact field value

The following query returns all users where the value of extendedPetsData matches the json variable exactly:

var json = [{ name: "Bob the dog" }, { name: "Claudine the cat" }];

const getUsers = await prisma.user.findMany({
  where: {
    extendedPetsData: {
      equals: json,
    },
  },
});

The following query returns all users where the value of extendedPetsData does not match the json variable exactly:

var json = [{ name: "Bob the dog" }, { name: "Claudine the cat" }];

const getUsers = await prisma.user.findMany({
  where: {
    extendedPetsData: {
      not: json,
    },
  },
});

You can also filter rows by the data inside a Json field. We call this advanced Json filtering. This functionality is supported by PostgreSQL and MySQL only with different syntaxes for the path option.

path syntax depending on database

The filters below use a path option to select specific parts of the Json value to filter on. The implementation of that filtering differs between connectors:

• The MySQL connector uses MySQL's implementation of JSON path
• The PostgreSQL connector uses the custom JSON functions and operators supported in version 12 and earlier

For example, the following is a valid MySQL path value:

$petFeatures.petName

The following is a valid PostgreSQL path value:

["petFeatures", "petName"]

Filter on object property

You can filter on a specific property inside a block of JSON. In the following examples, the value of extendedPetsData is a one-dimensional, unnested JSON object:

{
  "petName": "Claudine",
  "petType": "House cat"
}

The following query returns all users where the value of petName is "Claudine":

The following query returns all users where the value of petType contains "cat":

The following string filters are available:

• string_contains
• string_starts_with
• string_ends_with .

To use case insensitive filter with these, you can use the mode option:

Filter on nested object property

You can filter on nested JSON properties. In the following examples, the value of extendedPetsData is a JSON object with several levels of nesting.

{
  "pet1": {
    "petName": "Claudine",
    "petType": "House cat"
  },
  "pet2": {
    "petName": "Sunny",
    "petType": "Gerbil",
    "features": {
      "eyeColor": "Brown",
      "furColor": "White and black"
    }
  }
}

The following query returns all users where "pet2" → "petName" is "Sunny":

The following query returns all users where:

• "pet2" → "petName" is "Sunny"
• "pet2" → "features" → "furColor" contains "black"

Filtering on an array value

You can filter on the presence of a specific value in a scalar array (strings, integers). In the following example, the value of extendedPetsData is an array of strings:

["Claudine", "Sunny"]

The following query returns all users with a pet named "Claudine":

The following array filters are available:

• array_contains
• array_starts_with
• array_ends_with

Filtering on nested array value

You can filter on the presence of a specific value in a scalar array (strings, integers). In the following examples, the value of extendedPetsData includes nested scalar arrays of names:

{
  "cats": { "owned": ["Bob", "Sunny"], "fostering": ["Fido"] },
  "dogs": { "owned": ["Ella"], "fostering": ["Prince", "Empress"] }
}

Scalar value arrays

The following query returns all users that foster a cat named "Fido":

The following query returns all users that foster cats named "Fido" and "Bob":

JSON object arrays

• If you are using PostgreSQL, you must pass in an array of objects to match, even if that array only contains one object:

  [{ status: "expired", insuranceID: 92 }]
  // PostgreSQL
  

  If you are using MySQL, you must pass in a single object to match:

  { status: "expired", insuranceID: 92 }
  // MySQL
  

• If your filter array contains multiple objects, PostgreSQL will only return results if all objects are present - not if at least one object is present.

• You must set array_contains to a JSON object, not a string. If you use a string, Prisma Client escapes the quotation marks and the query will not return results. For example:

  array_contains: '[{"status": "expired", "insuranceID": 92}]';
  

  is sent to the database as:

  [{\"status\": \"expired\", \"insuranceID\": 92}]
  

Targeting an array element by index

You can filter on the value of an element in a specific position.

{ "owned": ["Bob", "Sunny"], "fostering": ["Fido"] }

Filtering on object key value inside array

Depending on your provider, you can filter on the key value of an object inside an array.

In the following example, the value of extendedPetsData is an array of objects with a nested insurances array, which contains two objects:

[
  {
    "petName": "Claudine",
    "petType": "House cat",
    "insurances": [
      { "insuranceID": 92, "status": "expired" },
      { "insuranceID": 12, "status": "active" }
    ]
  },
  {
    "petName": "Sunny",
    "petType": "Gerbil"
  },
  {
    "petName": "Gerald",
    "petType": "Corn snake"
  },
  {
    "petName": "Nanna",
    "petType": "Moose"
  }
]

The following query returns all users where at least one pet is a moose:

const getUsers = await prisma.user.findMany({
  where: {
    extendedPetsData: {
      path: "$[*].petType",
      array_contains: "Moose",
    },
  },
});

• $[*] is the root array of pet objects
• petType matches the petType key in any pet object

The following query returns all users where at least one pet has an expired insurance:

const getUsers = await prisma.user.findMany({
  where: {
    extendedPetsData: {
      path: "$[*].insurances[*].status",
      array_contains: "expired",
    },
  },
});

• $[*] is the root array of pet objects
• insurances[*] matches any insurances array inside any pet object
• status matches any status key in any insurance object

The following example assumes that the value of extendedPetsData is some variation of the following:

{
  "petName": "Claudine",
  "petType": "House cat",
  "insurances": [
    { "insuranceID": 92, "status": "expired" },
    { "insuranceID": 12, "status": "active" }
  ]
}

The following example:

1. Gets all users
2. Change the "status" of each insurance object to "expired"
3. Get all users that have an expired insurance where the ID is 92

There are two types of null values possible for a JSON field in an SQL database.

• Database NULL: The value in the database is a NULL.
• JSON null: The value in the database contains a JSON value that is null.

To differentiate between these possibilities, we've introduced three null enums you can use:

• JsonNull: Represents the null value in JSON.
• DbNull: Represents the NULL value in the database.
• AnyNull: Represents both null JSON values and NULL database values. (Only when filtering)

For example:

model Log {
  id   Int  @id
  meta Json
}

Here is an example of using AnyNull:

import { Prisma } from "@prisma/client";

prisma.log.findMany({
  where: {
    data: {
      meta: {
        equals: Prisma.AnyNull,
      },
    },
  },
});

Inserting null Values

This also applies to create, update and upsert. To insert a null value into a Json field, you would write:

import { Prisma } from "@prisma/client";

prisma.log.create({
  data: {
    meta: Prisma.JsonNull,
  },
});

And to insert a database NULL into a Json field, you would write:

import { Prisma } from "@prisma/client";

prisma.log.create({
  data: {
    meta: Prisma.DbNull,
  },
});

Filtering by null Values

To filter by JsonNull or DbNull, you would write:

import { Prisma } from "@prisma/client";

prisma.log.findMany({
  where: {
    meta: {
      equals: Prisma.AnyNull,
    },
  },
});

Prisma's Json fields are untyped by default. To add strong typing, you can use the external package prisma-json-types-generator.

1. First, install the package and add the generator to your schema.prisma:

   schema.prisma

   generator client {
     provider = "prisma-client"
     output   = "./generated"
   }
   
   generator json {
     provider = "prisma-json-types-generator"
   }
   

2. Next, link a field to a TypeScript type using an AST comment.

   schema.prisma

   model Log {
     id   Int @id
   
     /// [LogMetaType]
     meta Json
   }
   

3. Then, define LogMetaType in a type declaration file (e.g., types.ts) that is included in your tsconfig.json.

   types.ts

   declare global {
     namespace PrismaJson {
       type LogMetaType = { timestamp: number; host: string };
     }
   }
   
   // This file must be a module.
   export {};
   

Now, Log.meta will be strongly typed as { timestamp: number; host: string }.

Typing String Fields and Advanced Features

You can also apply these techniques to String fields. This is especially useful for creating string-based enums directly in your schema when your database does not support enum types.

model Post {
  id     Int    @id

  /// !['draft' | 'published']
  status String

  /// [LogMetaType]
  meta   Json[]
}

This results in post.status being strongly typed as 'draft' | 'published' and post.meta as LogMetaType[].

For a complete guide on configuration, monorepo setup, and other advanced features, please refer to the official prisma-json-types-generator documentation.

Can you select a subset of JSON key/values to return?

No - it is not yet possible to select which JSON elements to return. Prisma Client returns the entire JSON object.

Can you filter on the presence of a specific key?

No - it is not yet possible to filter on the presence of a specific key.

Is case insensitive filtering supported?

Yes - you can use the mode: 'insensitive' option with string filters like string_contains, string_starts_with, and string_ends_with. See Filter on object property for examples.

Can you sort an object property within a JSON value?

No, sorting object properties within a JSON value (order-by-prop) is not currently supported.

How to set a default value for JSON fields?

When you want to set a @default value the Json type, you need to enclose it with double-quotes inside the @default attribute (and potentially escape any "inner" double-quotes using a backslash), for example:

model User {
  id    Int  @id @default(autoincrement())
  json1 Json @default("[]")
  json2 Json @default("{ \"hello\": \"world\" }")
}