TypeScript icon, indicating that this package has built-in type declarations

3.0.0-alpha.7 • Public • Published


Created by Colin McDonnell License npm stars coverage

if you're happy and you know it, star this repo

Migrating from v1

If you're upgrading straight to v3 from v1, you'll need to be aware of the breaking changes introduced in both v2 and v3. The v1->v2 migration guide is here.

Migrating from v2

You can install v3 with zod@next. (v2 will continue to be availabe with zod@beta for the time being.)

npm install zod@next
yarn add zod@next

Breaking changes in v3

  • The minimum TypeScript version has increased from 3.7 to 4.1. Several features have been rewritten to use recursive conditional types, an incredibly powerful new feature introduced in TS4.1.

  • Transformer syntax. Previously transformers required an input, an output schema, and a function to tranform between them. You created transformers like z.transform(A, B, func), where A and B are Zod schemas. This is no longer the case. Accordingly:

    ⚠️ The old syntax (z.transformer(A, B, func)) is no longer available. ⚠️ The convenience method A.transform(B, func) is no longer available.

    Instead, you apply transformations by simply using the .transform() method that exists on all Zod schema. For example: z.string().transform(val => val.length).

    Importantly, transformations are now interleaved with refinements. So you can build chains of refinement + transformation logic that are executed in sequence:

    const test = z
      .transform((val) => val.length)
      .refine((val) => val > 5, { message: "Input is too short" })
      .transform((val) => val * 2);
    test.parse("12characters"); // => 24
  • Type guards (the .check() method) have been removed. Type guards interact with transformers in unintuitive ways so they were removed. Use .safeParse instead.

What is Zod

Zod is a TypeScript-first schema declaration and validation library. I'm using the term "schema" to broadly refer to any data type/structure, from a simple string to a complex nested object.

Zod is designed to be as developer-friendly as possible. My goal is to eliminate duplicative type declarations wherever possible. With Zod, you declare a validator once and Zod will automatically infer the static TypeScript type. It's easy to compose simpler types into complex data structures.

Some other great aspects:

  • Zero dependencies
  • Plain JavaScript: works in browsers and Node.js
  • Tiny: 8kb minified + zipped
  • Immutability: methods (i.e. .optional() return a new instance
  • Concise, chainable interface
  • Functional approach: parse, don't validate
  • Works with plain JavaScript too! You don't need to use TypeScript.


I work on Zod in my free time, so if you're making money from a product that is built with Zod, I'd massively appreciate sponsorship at any level. For solo devs, I recommend the Chipotle Bowl tier or the Cup of Coffee tier. If you're making money from a product you built using Zod, consider the Startup tier. You can learn more about the tiers at github.com/sponsors/colinhacks.


Kevin Simper

Brandon Bayer
@flybayer, creator of Blitz.js

Bamboo Creative

To get your name + Twitter + website here, sponsor Zod at the Freelancer or Consultancy tier.

Table of contents


To use the beta of Zod 2 (recommended for new projects).

yarn add zod@beta
npm install zod@beta

To install the most recent v1 version:

yarn add zod
npm install zod

TypeScript requirements

  • Zod 3.x requires TypeScript 3.8+
  • Zod 2.x requires TypeScript 3.7+
  • Zod 1.x requires TypeScript 3.3+

Support for TS 3.2 was dropped with the release of zod@1.10 on 19 July 2020

You must enable strictNullChecks or use strict mode which includes strictNullChecks . Otherwise Zod can't correctly infer the types of your schemas!

// tsconfig.json
  // ...
  "compilerOptions": {
    // ...
    "strictNullChecks": true


Zod is a validation library designed for optimal developer experience. It's a TypeScript-first schema declaration library with rigorous inferred types, incredible developer experience, and a few killer features missing from the existing libraries.

  • Zero dependencies (5kb compressed)
  • Immutability; methods (i.e. .optional() return a new instance
  • Concise, chainable interface
  • Functional approach ("Parse, don't validate!")


You can create a Zod schema for any TypeScript primitive.

import * as z from "zod";

// primitive values

// empty types

// catch-all types


const tuna = z.literal("tuna");
const twelve = z.literal(12);
const tru = z.literal(true);

Currently there is no support for Date or bigint literals in Zod. If you have a use case for this feature, please file an issue.



.parse(data:unknown): T

Given any Zod schema, you can call its .parse method to check data is valid. If it is, a value is returned with full type information! Otherwise, an error is thrown.

IMPORTANT: In Zod 2 and Zod 1.11+, the value returned by .parse is a deep clone of the variable you passed in. This was also the case in zod@1.4 and earlier.

const stringSchema = z.string();
stringSchema.parse("fish"); // => returns "fish"
stringSchema.parse(12); // throws Error('Non-string type: number');

Safe parse

.safeParse(data:unknown): { success: true; data: T; } | { success: false; error: ZodError; }

If you don't want Zod to throw when validation errors occur, you can use .safeParse . This method returns an object, even if validation errors occur:

// => { success: false; error: ZodError }

// => { success: true; data: 'billie' }

There is also an asynchronous version:

await stringSchema.safeParseAsync("billie");

You must use .parseAsync() or .safeParseAsync() if your schema contains asynchronous refinements for transformers.

The result is a discriminated union so you can handle errors very conveniently:

const result = stringSchema.safeParse("billie");
if (!result.success) {
  // handle error then return

// underneath the if statement, TypeScript knows
// that validation passed

Errors thrown from within refinement functions will not be caught.

Type guards


⚠️ Type guards have been removed in Zod 3 because of compatibility issues with transformers. You should safeParse to achieve a similar result that is compatible with transformers.


⚠️ Refinements must not throw. Instead they should return a falsy value to signal failure.

.refine(validator: (data:T)=>any, params?: RefineParams)

Zod lets you provide custom validation logic via refinements.

Zod was designed to mirror TypeScript as closely as possible. But there are many so-called "refinement types" you may wish to check for that can't be represented in TypeScript's type system. For instance: checking that a number is an integer or that a string is a valid email address.

For example, you can define a custom validation check on any Zod schema with .refine :

const myString = z.string().refine((val) => val.length <= 255, {
  message: "String can't be more than 255 characters",

Refinements can also be async:

const userId = z.string().refine(async (id) => {
  // verify that ID exists in database
  return true;

⚠️If you use async refinements, you must use the .parseAsync method to parse data! Otherwise Zod will throw an error.

As you can see, .refine takes two arguments.

  1. The first is the validation function. This function takes one input (of type T — the inferred type of the schema) and returns any. Any truthy value will pass validation. (Prior to zod@1.6.2 the validation function had to return a boolean.)
  2. The second argument accepts some options. You can use this to customize certain error-handling behavior:
type RefineParams = {
  // override error message
  message?: string;

  // appended to error path
  path?: (string | number)[];

  // params object you can use to customize message
  // in error map
  params?: object;

These options let you define powerful custom behavior. Zod is commonly used for form validation. If you want to verify that "password" and "confirm" match, you can do so like this:

const passwordForm = z
    password: z.string(),
    confirm: z.string(),
  .refine((data) => data.password === data.confirm, {
    message: "Passwords don't match",
    path: ["confirm"], // path of error
  .parse({ password: "asdf", confirm: "qwer" });

Because you provided a path parameter, the resulting error will be:

ZodError {
  issues: [{
    "code": "custom",
    "path": [ "confirm" ],
    "message": "Passwords don't match"

Note that the path is set to ["confirm"] , so you can easily display this error underneath the "Confirm password" textbox.

Important note, the value passed to the path option is concatenated to the actual error path. So if you took passwordForm from above and nested it inside another object, you would still get the error path you expect.

const allForms = z.object({ passwordForm }).parse({
  passwordForm: {
    password: "asdf",
    confirm: "qwer",

would result in

ZodError {
  issues: [{
    "code": "custom",
    "path": [ "passwordForm", "confirm" ],
    "message": "Passwords don't match"

Type inference

You can extract the TypeScript type of any schema with z.infer<typeof mySchema> .

const A = z.string();
type A = z.infer<typeof A>; // string

const u: A = 12; // TypeError
const u: A = "asdf"; // compiles

We'll include examples of inferred types throughout the rest of the documentation.


There are a handful of string-specific validations.

All of these validations allow you to optionally specify a custom error message.


Use the .nonempty method if you want the empty string ( "" ) to be considered invalid.

Check out validator.js for a bunch of other useful string validation functions.

Custom error messages

Like .refine , The final (optional) argument is an object that lets you provide a custom error in the message field.

z.string().min(5, { message: "Must be 5 or more characters long" });
z.string().max(5, { message: "Must be 5 or fewer characters long" });
z.string().length(5, { message: "Must be exactly 5 characters long" });
z.string().email({ message: "Invalid email address." });
z.string().url({ message: "Invalid url" });
z.string().uuid({ message: "Invalid UUID" });

To see the email and url regexes, check out this file. To use a more advanced method, use a custom refinement.


There are a handful of number-specific validations.


z.number().int(); // value must be an integer

z.number().positive(); //     > 0
z.number().nonnegative(); //  >= 0
z.number().negative(); //     < 0
z.number().nonpositive(); //  <= 0

You can optionally pass in a params object as the second argument to provide a custom error message.

z.number().max(5, { message: "this👏is👏too👏big" });


// all properties are required by default
const dogSchema = z.object({
  name: z.string(),
  age: z.number(),

type Dog = z.infer<typeof dogSchema>;
equivalent to:
type Dog = { 
  name: string; 
  age: number;

const cujo = dogSchema.parse({
  name: "Cujo",
  age: 4,
}); // passes, returns Dog

const fido: Dog = {
  name: "Fido",
}; // TypeError: missing required property `age`

.shape property

Use .shape to access an object schema's property schemas.

const Location = z.object({
  latitude: z.number(),
  longitude: z.number(),

const Business = z.object({
  location: Location,

Business.shape.location; // => Location schema


You can combine two object schemas with .merge , like so:

const BaseTeacher = z.object({ subjects: z.array(z.string()) });
const HasID = z.object({ id: z.string() });

const Teacher = BaseTeacher.merge(HasID);
type Teacher = z.infer<typeof Teacher>; // => { subjects: string[], id: string }

You're able to fluently chain together many .merge calls as well:

// chaining mixins
const Teacher = BaseTeacher.merge(HasId).merge(HasName).merge(HasAddress);

IMPORTANT: the schema returned by .merge is the intersection of the two schemas. The schema passed into .merge does not "overwrite" properties of the original schema. To demonstrate:

const Obj1 = z.object({ field: z.string() });
const Obj2 = z.object({ field: z.number() });

const Merged = Obj1.merge(Obj2);

type Merged = z.infer<typeof merged>;
// => { field: never }
// because no type can simultaneously be both a string and a number

To "overwrite" existing keys, use .extend (documented below).

Extending objects

You can add additional fields an object schema with the .extend method.

Before zod@1.8 this method was called .augment . The augment method is still available for backwards compatibility but it is deprecated and will be removed in a future release.

const Animal = z
    species: z.string(),
    population: z.number(),

⚠️ You can use .extend to overwrite fields! Be careful with this power!

// overwrites `species`
const ModifiedAnimal = Animal.extend({
  species: z.array(z.string()),

// => { population: number, species: string[] }

Pick and omit

Object masking is one of Zod's killer features. It lets you create slight variations of your object schemas easily and succinctly. Inspired by TypeScript's built-in Pick and Omit utility types, all Zod object schemas have .pick and .omit methods that return a "masked" version of the schema.

const Recipe = z.object({
  id: z.string(),
  name: z.string(),
  ingredients: z.array(z.string()),

To only keep certain keys, use .pick .

const JustTheName = Recipe.pick({ name: true });

type JustTheName = z.infer<typeof JustTheName>;
// => { name: string }

To remove certain keys, use .omit .

const NoIDRecipe = Recipe.omit({ id: true });

type NoIDRecipe = z.infer<typeof NoIDRecipe>;
// => { name: string, ingredients: string[] }

This is useful for database logic, where endpoints often accept as input slightly modified versions of your database schemas. For instance, the input to a hypothetical createRecipe endpoint would accept the NoIDRecipe type, since the ID will be generated by your database automatically.

This is a vital feature for implementing typesafe backend logic, yet as far as I know, no other validation library (yup, Joi, io-ts, runtypes, class-validator, ow...) offers similar functionality as of this writing (April 2020). This is one of the must-have features that inspired the creation of Zod.


Inspired by the built-in TypeScript utility type Partial, all Zod object schemas have a .partial method that makes all properties optional.

Starting from this object:

const user = z.object({
  username: z.string(),
  location: z.object({
    latitude: z.number(),
    longitude: z.number(),
  { username: string, location: { city: number, state: number } }

We can create a partial version:

const partialUser = user.partial();
  username?: string | undefined,
  location?: {
    city: number;
    state: number;
  } | undefined

// equivalent to:
const partialUser = z.object({
  username: user.shape.username.optional(),
  location: user.shape.location.optional(),

Or you can use .deepPartial :

const deepPartialUser = user.deepPartial();

  username?: string | undefined, 
  location?: {
    latitude?: number | undefined;
    longitude?: number | undefined;
  } | undefined

Important limitation: deep partials only work as expected in hierarchies of object schemas. It also can't be used on recursive schemas currently, since creating a recursive schema requires casting to the generic ZodSchema type (which doesn't include all the methods of the ZodObject class). Currently an improved version of Zod is under development that will have better support for recursive schemas.

Unknown keys

By default Zod object schema strip unknown keys from the output.

⚠️ Before version 2, Zod did NOT allow unknown keys by default.

Zod will return

const person = z.object({
  name: z.string(),

  name: "bob dylan",
  extraKey: 61,
// => { name: "bob dylan" }

Pass through unknown keys

If you want to pass through unknown keys, use .passthrough() .

For backwards compatibility, you can also use .nonstrict() which behaves identically.

const person = z
    name: z.string(),

  name: "bob dylan",
  extraKey: 61,
// => { name: "bob dylan", extraKey: 61 }

Disallow unknown keys

You can disallow unknown keys with .strict() . If there are any unknown keys in the input, Zod will throw an error.

const person = z
    name: z.string(),

  name: "bob dylan",
  extraKey: 61,
// => throws ZodError

Primitives and nonprimitives

Zod provides a convenience method for automatically picking all primitive or non-primitive fields from an object schema.

const Post = z.object({
  title: z.string()

const User = z.object({
  id: z.number(),
  name: z.string(),
  posts: z.array(Post)

const UserFields = User.primitives();
typeof UserFields = z.infer<typeof UserFields>;
// => { id: number; name; string; }

const UserRelations = User.nonprimitives();
typeof UserFields = z.infer<typeof UserFields>;
// => { posts: Post[] }

These schemas are considering "primitive":

  • string
  • number
  • boolean
  • bigint
  • date
  • null/undefined
  • enums
  • any array of the above types
  • any union of the above types


You can add a catchall schema with .catchall() . All unknown keys will be validated against the catchall schema.

const person = z
    name: z.string(),

  name: "bob dylan",
  validExtraKey: 61, // works fine
// => { name: "bob dylan", validExtraKey: 61 }

Using .catchall() overrides .passsthrough() , .strip() , or .strict() . All keys are now considered "known".


Record schemas are used to validate types such as this:

type NumberCache = { [k: string]: number };

If you want to validate that all the values of an object match some schema, without caring about the keys, you should use a Record.

const User = z.object({
  name: z.string(),

const UserStore = z.record(User);

type UserStore = z.infer<typeof UserStore>;
// => { [k: string]: User }

This is particularly useful for storing or caching items by ID.

const userStore: UserStore = {};

userStore["77d2586b-9e8e-4ecf-8b21-ea7e0530eadd"] = {
  name: "Carlotta",
}; // passes

userStore["77d2586b-9e8e-4ecf-8b21-ea7e0530eadd"] = {
  whatever: "Ice cream sundae",
}; // TypeError

And of course you can call .parse just like any other Zod schema.

  user_1328741234: { name: "James" },
}); // => passes

A note on numerical keys

You may have expected z.record() to accept two arguments, one for the keys and one for the values. After all, TypeScript's built-in Record type does: Record<KeyType, ValueType> . Otherwise, how do you represent the TypeScript type Record<number, any> in Zod?

As it turns out, TypeScript's behavior surrounding [k: number] is a little unintuitive:

const testMap: { [k: number]: string } = {
  1: "one",

for (const key in testMap) {
  console.log(`${key}: ${typeof key}`);
// prints: `1: string`

As you can see, JavaScript automatically casts all object keys to strings under the hood.

Since Zod is trying to bridge the gap between static and runtime types, it doesn't make sense to provide a way of creating a record schema with numerical keys, since there's no such thing as a numerical key in runtime JavaScript.


There are two ways to define array schemas:

z.array(arg: ZodSchema)

First, you can create an array schema with the z.array() function; it accepts another ZodSchema, which defines the type of each array element.

const stringArray = z.array(z.string());
// inferred type: string[]

the .array() method

Second, you can call the .array() method on any Zod schema:

const stringArray = z.string().array();
// inferred type: string[]

You have to be careful with the .array() method. It returns a new ZodArray instance. This means you need to be careful about the order in which you call methods. These two schemas are very different:

z.string().undefined().array(); // (string | undefined)[]
z.string().array().undefined(); // string[] | undefined

Non-empty lists

const nonEmptyStrings = z.string().array().nonempty();
// [string, ...string[]]

nonEmptyStrings.parse([]); // throws: "Array cannot be empty"
nonEmptyStrings.parse(["Ariana Grande"]); // passes

Length validations

// must contain 5 or more items

// must contain 5 or fewer items

// must contain exactly 5 items


Zod includes a built-in z.union method for composing "OR" types.

const stringOrNumber = z.union([z.string(), z.number()]);

stringOrNumber.parse("foo"); // passes
stringOrNumber.parse(14); // passes

Zod will test the input against each of the "options" in order and return the first value that validates successfully.

Optional types

You can make any schema optional with z.optional():

const A = z.optional(z.string());

A.parse(undefined); // => passes, returns undefined
type A = z.infer<typeof A>; // string | undefined

You can also call the .optional() method on an existing schema:

const B = z.boolean().optional();

const C = z.object({
  username: z.string().optional(),
type C = z.infer<typeof C>; // { username?: string | undefined };

Nullable types

Similarly, you can create nullable types like so:

const D = z.nullable(z.string());
D.parse("asdf"); // => "asdf"
D.parse(null); // => null

Or you can use the .nullable() method on any existing schema:

const E = z.string().nullable(); // equivalent to D
type E = z.infer<typeof D>; // string | null

You can create unions of any two or more schemas.


There are two ways to define enums in Zod.

Zod enums

An enum is just a union of string literals, so you could define an enum like this:

const FishEnum = z.union([

FishEnum.parse("Salmon"); // => "Salmon"
FishEnum.parse("Flounder"); // => throws

For convenience Zod provides a built-in z.enum() function. Here's is the equivalent code:

const FishEnum = z.enum(["Salmon", "Tuna", "Trout"]);

type FishEnum = z.infer<typeof FishEnum>;
// 'Salmon' | 'Tuna' | 'Trout'

Important! You need to pass the literal array directly into z.enum(). Do not define it separately, than pass it in as a variable! This is required for proper type inference.


To get autocompletion with a Zod enum, use the .enum property of your schema:

FishEnum.enum.Salmon; // => autocompletes

=> {
  Salmon: "Salmon",
  Tuna: "Tuna",
  Trout: "Trout",

You can also retrieve the list of options as a tuple with the .options property:

FishEnum.options; // ["Salmon", "Tuna", "Trout"]);

Native enums

⚠️ nativeEnum() requires TypeScript 3.6 or higher!

Zod enums are the recommended approach to defining and validating enums. But there may be scenarios where you need to validate against an enum from a third-party library, or perhaps you don't want to rewrite your existing enums. For this you can use z.nativeEnum() .

Numeric enums

enum Fruits {

const FruitEnum = z.nativeEnum(Fruits);
type FruitEnum = z.infer<typeof FruitEnum>; // Fruits

FruitEnum.parse(Fruits.Apple); // passes
FruitEnum.parse(Fruits.Banana); // passes
FruitEnum.parse(0); // passes
FruitEnum.parse(1); // passes
FruitEnum.parse(3); // fails

String enums

enum Fruits {
  Apple = "apple",
  Banana = "banana",
  Cantaloupe, // you can mix numerical and string enums

const FruitEnum = z.nativeEnum(Fruits);
type FruitEnum = z.infer<typeof FruitEnum>; // Fruits

FruitEnum.parse(Fruits.Apple); // passes
FruitEnum.parse(Fruits.Cantaloupe); // passes
FruitEnum.parse("apple"); // passes
FruitEnum.parse("banana"); // passes
FruitEnum.parse(0); // passes
FruitEnum.parse("Cantaloupe"); // fails

Const enums

The .nativeEnum() function works for as const objects as well. ⚠️ as const required TypeScript 3.4+!

const Fruits = {
  Apple: "apple",
  Banana: "banana",
  Cantaloupe: 3,
} as const;

const FruitEnum = z.nativeEnum(Fruits);
type FruitEnum = z.infer<typeof FruitEnum>; // "apple" | "banana" | 3

FruitEnum.parse("apple"); // passes
FruitEnum.parse("banana"); // passes
FruitEnum.parse(3); // passes
FruitEnum.parse("Cantaloupe"); // fails


Intersections are useful for creating "logical AND" types.

const a = z.union([z.number(), z.string()]);
const b = z.union([z.number(), z.boolean()]);
const c = z.intersection(a, b);

type c = z.infer<typeof c>; // => number

const stringAndNumber = z.intersection(z.string(), z.number());
type Never = z.infer<typeof stringAndNumber>; // => never


These differ from arrays in that they have a fixed number of elements, and each element can have a different type.

const athleteSchema = z.tuple([
  // takes an array of schemas
  z.string(), // name
  z.number(), // jersey number
    pointsScored: z.number(),
  }), // statistics

type Athlete = z.infer<typeof athleteSchema>;
// type Athlete = [string, number, { pointsScored: number }]

Recursive types

You can define a recursive schema in Zod, but because of a limitation of TypeScript, their type can't be statically inferred. If you need a recursive Zod schema you'll need to define the type definition manually, and provide it to Zod as a "type hint".

interface Category {
  name: string;
  subcategories: Category[];

const Category: z.ZodSchema<Category> = z.lazy(() =>
    name: z.string(),
    subcategories: z.array(Category),

  name: "People",
  subcategories: [
      name: "Politicians",
      subcategories: [{ name: "Presidents", subcategories: [] }],
}); // passes

Unfortunately this code is a bit duplicative, since you're declaring the types twice: once in the interface and again in the Zod definition.

If your schema has lots of primitive fields, there's a way of reducing the amount of duplication:

// define all the non-recursive stuff here
const BaseCategory = z.object({
  name: z.string(),
  tags: z.array(z.string()),
  itemCount: z.number(),

// create an interface that extends the base schema
interface Category extends z.infer<typeof BaseCategory> {
  subcategories: Category[];

// merge the base schema with
// a new Zod schema containing relations
const Category: z.ZodSchema<Category> = BaseCategory.merge(
    subcategories: z.lazy(() => z.array(Category)),

JSON type

If you want to validate any JSON value, you can use the snippet below. This requires TypeScript 3.7 or higher!

type Literal = boolean | null | number | string;
type Json = Literal | { [key: string]: Json } | Json[];
const literalSchema = z.union([z.string(), z.number(), z.boolean(), z.null()]);
const jsonSchema: z.ZodSchema<Json> = z.lazy(() =>
  z.union([literalSchema, z.array(jsonSchema), z.record(jsonSchema)])

  // data

Thanks to ggoodman for suggesting this.

Cyclical objects

As of Zod 2, Zod no longer supports cyclical objects. If you absolutely need this feature you can still use Zod v1.


const numberPromise = z.promise(z.number());

"Parsing" works a little differently with promise schemas. Validation happens in two parts:

  1. Zod synchronously checks that the input is an instance of Promise (i.e. an object with .then and .catch methods.).
  2. Zod waits for the promise to resolve then validates the resolved value.
// ZodError: Non-Promise type: string

// => Promise<number>

const test = async () => {
  await numberPromise.parse(Promise.resolve("tuna"));
  // ZodError: Non-number type: string

  await numberPromise.parse(Promise.resolve(3.14));
  // => 3.14

Non-native promise implementations

When "parsing" a promise, Zod checks that the passed value is an object with .then and .catch methods — that's it. So you should be able to pass non-native Promises (Bluebird, etc) into z.promise(...).parse with no trouble. One gotcha: the return type of the parse function will be a native Promise , so if you have downstream logic that uses non-standard Promise methods, this won't work.


const stringNumberMap = z.map(z.string(), z.number());

type StringNumberMap = z.infer<typeof stringNumberMap>;
// type StringNumber = Map<string, number>


You can use z.instanceof to create a schema that checks if the input is an instance of a class.

class Test {
  name: string;

const TestSchema = z.instanceof(Test);

const blob: any = "whatever";
const parsed = TestSchema.safeParse(blob);
if (parsed.success) {
  parsed.data; // instance of Test

Function schemas

Zod also lets you define "function schemas". This makes it easy to validate the inputs and outputs of a function without intermixing your validation code and "business logic".

You can create a function schema with z.function(args, returnType) .

const myFunction = z.function();

type myFunction = z.infer<typeof myFunction>;
// => ()=>unknown

You can use the .args and .returns methods to refine your function schema:

const myFunction = z
  .args(z.string(), z.number()) // accepts an arbitrary number of arguments
type myFunction = z.infer<typeof myFunction>;
// => (arg0: string, arg1: number)=>boolean

You can use the special z.void() option if your function doesn't return anything. This will let Zod properly infer the type of void-returning functions. (Void-returning function can actually return either undefined or null.)

Function schemas have an .implement() method which accepts a function and returns a new function.

const trimmedLength = z
  .args(z.string()) // accepts an arbitrary number of arguments
  .implement((x) => {
    // TypeScript knows x is a string!
    return x.trim().length;

trimmedLength("sandwich"); // => 8
trimmedLength(" asdf "); // => 4

trimmedLength now automatically validates both its inputs and return value against the schemas provided to z.function . If either is invalid, the function throws. This way you can confidently write application logic in a "validated function" without worrying about invalid inputs, scattering schema.validate() calls in your endpoint definitions, or writing duplicative types for your functions.

Here's a more complex example showing how to write a typesafe API query endpoint:

const FetcherEndpoint = z
  .function(args, returnType)
  .args(z.object({ id: z.string() }))
        id: string(),
        name: string(),

const getUserByID = FetcherEndpoint.validate(async (args) => {
  args; // => { id: string }

  const user = await User.findByID(args.id);

  // TypeScript statically verifies that value returned by
  // this function is of type Promise<{ id: string; name: string; }>
  return "salmon"; // TypeError

  return user; // compiles successfully

This is particularly useful for defining HTTP or RPC endpoints that accept complex payloads that require validation. Moreover, you can define your endpoints once with Zod and share the code with both your client and server code to achieve end-to-end type safety.

// Express example
server.get(`/user/:id`, async (req, res) => {
  const user = await getUserByID({ id: req.params.id }).catch((err) => {



You can integrate custom data transformations into your schemas with transformers. Transformers are just another type of Zod schema.

⚠️ Transformation functions must not throw. Make sure to use refinements before the transformer to make sure the input can be parsed by the transformer.

⚠️ The z.transformer(A, B, func) syntax was removed in Zod 3.


You use transformers using the .transform() method.

const stringToNumber = z.string().transform((val) => myString.length);
stringToNumber.parse("string"); // => 6

Chaining order

Note that stringToNumber above is an instance of the ZodTransformer subclass. It is NOT an instance of ZodString. If you want to use the built-in methods of ZodString (e.g. .email()) you must apply those methods before any transformations.

const emailToDomain = z
  .transform((val) => val.split("@")[1]);

emailToDomain.parse("colinhacks@example.com"); // => example.com

Async transformations

Transformations can also be async.

const IdToUser = z.transformer(
  (userId) => async (id) => {
    return await getUserById(id);

⚠️ If your schema contains asynchronous transformers, you must use .parseAsync() or .safeParseAsync() to parse data. Otherwise Zod will throw an error.

Default values

You can use transformers to implement the concept of "default values" in Zod.

const stringWithDefault = z
  .transform((val) => {
    return typeof val !== "undefined" ? val : "tuna";

stringWithDefault.parse(undefined); // => "tuna"


You can also use the provided convenience method .default like so:

const stringWithDefault = z.string().default("tuna");

Optionally, you can pass a function into .default that will be re-executed whenever a default value needs to be generated:

const numberWithRandomDefault = z.number().default(Math.random);

numberWithRandomDefault.parse(undefined); // => 0.4413456736055323
numberWithRandomDefault.parse(undefined); // => 0.1871840107401901
numberWithRandomDefault.parse(undefined); // => 0.7223408162401552

Type inference for transformers

Every Zod schema is associated with an input type and and output type. For most schemas (e.g. z.string()) these two are the same. But for Transformers, they are different. For instance z.string().transform(val => val.length) has an input of string and an output of number.

Normally you can use z.infer<typeof A> to extract the inferred type of schema A. But for schemas we introduce two new methods that allow you to separately extract the input and output independently.

const stringToNumber = z.string().transform(val => val.length)

// it is equivalent to z.output<>
type out = z.output<stringToNumber>; // number

// you can use z.input<> to get the input type
type in = z.input<stringToNumber>; // string

// ⚠️ Important: z.infer gives the OUTPUT type!
type type = z.infer<stringToNumber>; // number


There is a dedicated guide on Zod's error handling system here: ERROR_HANDLING.md


There are a handful of other widely-used validation libraries, but all of them have certain design limitations that make for a non-ideal developer experience.



Doesn't support static type inference 😕



Yup is a full-featured library that was implemented first in vanilla JS, and later rewritten in TypeScript.


  • Supports for casting and transformation
  • All object fields are optional by default
  • Missing object methods: (partial, deepPartial)
  • Missing promise schemas
  • Missing function schemas
  • Missing union & intersection schemas



io-ts is an excellent library by gcanti. The API of io-ts heavily inspired the design of Zod.

In our experience, io-ts prioritizes functional programming purity over developer experience in many cases. This is a valid and admirable design goal, but it makes io-ts particularly hard to integrate into an existing codebase with a more procedural or object-oriented bias. For instance, consider how to define an object with optional properties in io-ts:

import * as t from "io-ts";

const A = t.type({
  foo: t.string,

const B = t.partial({
  bar: t.number,

const C = t.intersection([A, B]);

type C = t.TypeOf<typeof C>;
// returns { foo: string; bar?: number | undefined }

You must define the required and optional props in separate object validators, pass the optionals through t.partial (which marks all properties as optional), then combine them with t.intersection .

Consider the equivalent in Zod:

const C = z.object({
  foo: z.string(),
  bar: z.number().optional(),

type C = z.infer<typeof C>;
// returns { foo: string; bar?: number | undefined }

This more declarative API makes schema definitions vastly more concise.

io-ts also requires the use of gcanti's functional programming library fp-ts to parse results and handle errors. This is another fantastic resource for developers looking to keep their codebase strictly functional. But depending on fp-ts necessarily comes with a lot of intellectual overhead; a developer has to be familiar with functional programming concepts and the fp-ts nomenclature to use the library.

  • Supports codecs with serialization & deserialization transforms
  • Supports branded types
  • Supports advanced functional programming, higher-kinded types, fp-ts compatibility
  • Missing object methods: (pick, omit, partial, deepPartial, merge, extend)
  • Missing nonempty arrays with proper typing ([T, ...T[]])
  • Missing lazy/recursive types
  • Missing promise schemas
  • Missing function schemas
  • Missing union & intersection schemas
  • Missing support for parsing cyclical data (maybe)
  • Missing error customization



Good type inference support, but limited options for object type masking (no .pick , .omit , .extend , etc.). No support for Record s (their Record is equivalent to Zod's object ). They DO support branded and readonly types, which Zod does not.

  • Supports "pattern matching": computed properties that distribute over unions
  • Supports readonly types
  • Missing object methods: (pick, omit, partial, deepPartial, merge, extend)
  • Missing nonempty arrays with proper typing ([T, ...T[]])
  • Missing lazy/recursive types
  • Missing promise schemas
  • Missing union & intersection schemas
  • Missing error customization
  • Missing record schemas (their "record" is equivalent to Zod "object")



Ow is focused on function input validation. It's a library that makes it easy to express complicated assert statements, but it doesn't let you parse untyped data. They support a much wider variety of types; Zod has a nearly one-to-one mapping with TypeScript's type system, whereas ow lets you validate several highly-specific types out of the box (e.g. int32Array , see full list in their README).

If you want to validate function inputs, use function schemas in Zod! It's a much simpler approach that lets you reuse a function type declaration without repeating yourself (namely, copy-pasting a bunch of ow assertions at the beginning of every function). Also Zod lets you validate your return types as well, so you can be sure there won't be any unexpected data passed downstream.


View the changelog at CHANGELOG.md

Package Sidebar


Weekly Downloads






Unpacked Size

497 kB

Total Files


Last publish


  • jacksteamdev