ArkType
Overview
ArkType is a runtime validation library that uses TypeScript's own syntax for type definitions. Instead of learning a new API (z.string().email()), you write types the way you already know ("string.email"). Its own benchmarks put it far ahead of Zod on speed (the project cites an order of magnitude or more; measure your own payloads), and it gives readable error messages and a 1:1 match between your types and validators. Current release at the time of writing: 2.2.x. It implements Standard Schema, so it plugs into libraries that accept that interface.
When to Use
- Runtime validation where performance matters (hot paths, large payloads)
- Want to write validators using TypeScript syntax (not a builder API)
- Need detailed, human-readable error messages
- Replacing Zod in performance-critical applications
- Type-safe data transformations (morphs)
Instructions
Setup
npm install arktype
Basic Types
// types.ts — Define types using TypeScript syntax
import { type } from "arktype";
// String with constraints
const email = type("string.email");
const result = email("maya.chen@northwind.io"); // "maya.chen@northwind.io"
const error = email("not-an-email"); // ArkErrors: must be an email address (was "not-an-email")
// Object types — looks like TypeScript
const User = type({
name: "string >= 2", // String with min length 2
email: "string.email",
age: "number.integer >= 13", // Integer, min 13
role: "'admin' | 'user'", // Literal union
"bio?": "string <= 500", // Optional, max 500 chars
});
// Infer the TypeScript type — no separate interface needed
type User = typeof User.infer;
// { name: string; email: string; age: number; role: "admin" | "user"; bio?: string }
// Validate
const valid = User({
name: "Kai",
email: "kai.berg@northwind.io",
age: 25,
role: "user",
});
if (valid instanceof type.errors) {
console.log(valid.summary); // Human-readable error message
} else {
console.log(valid.name); // Typed as User
}
Arrays and Nested Types
// complex.ts — Complex nested types
import { type } from "arktype";
const Address = type({
street: "string",
city: "string",
zip: "string.numeric", // Only checks the string looks numeric; stays a string
country: "string == 2", // Exactly 2 characters (ISO code)
});
const Order = type({
id: "string.uuid",
items: type({
productId: "string",
quantity: "number.integer > 0",
price: "number > 0",
}).array(), // Array of items
shippingAddress: Address,
total: "number > 0",
status: "'pending' | 'shipped' | 'delivered'",
createdAt: "Date",
});
type Order = typeof Order.infer;
Morphs (Transforms)
string.trim, string.numeric.parse, string.json.parse and string.date.parse are built-in morphs: they validate and convert in one step. Plain string.numeric only validates and leaves a string.
// morphs.ts — Transform data during validation
import { type } from "arktype";
// Parse string to number
const numericString = type("string.numeric.parse");
numericString("42"); // 42 (number)
// Defaults and custom checks
const Settings = type({ retries: "number.integer = 3" });
Settings({}); // { retries: 3 }
const Even = type("number").narrow((n, ctx) => n % 2 === 0 || ctx.mustBe("even"));
Even(3).summary; // "must be even (was 3)"
// Parse and transform API input
const CreateUserInput = type({
name: "string.trim", // Auto-trim
email: type("string.email").pipe((e) => e.toLowerCase()), // Lowercase
age: type("string.numeric.parse"), // String → number
tags: type("string").pipe((s) => s.split(",")), // "a,b,c" → ["a","b","c"]
});
Scopes (Reusable Type Systems)
// scope.ts — Define interconnected types
import { scope } from "arktype";
const types = scope({
user: {
id: "string.uuid",
name: "string >= 2",
email: "string.email",
"posts?": "post[]",
},
post: {
id: "string.uuid",
title: "string >= 1",
content: "string",
"author?": "user", // Cyclic reference (optional so data can terminate)
tags: "string[]",
},
}).export();
const out = types.user({ id: crypto.randomUUID(), name: "Kai" });
if (out instanceof type.errors) console.error(out.summary);
Throwing, JSON Schema and Standard Schema
const Config = type({ port: "number.integer", "host?": "string" });
const config = Config.assert({ port: 8080 }); // returns data or throws TraversalError
const schema = Config.toJsonSchema(); // JSON Schema draft 2020-12 object
Config["~standard"].vendor; // "arktype" (Standard Schema interface)
Examples
Example 1: Validate API request bodies
User prompt: "Validate incoming POST requests in my Express API with clear error messages."
The agent installs arktype, defines one type({...}) per endpoint body, and writes a middleware that calls the schema and checks out instanceof type.errors. On failure it responds 400 with out.summary; on success it passes the typed data on. A bad body such as { "age": 3 } returns age must be at least 13 (was 3).
Example 2: Replace Zod with ArkType
User prompt: "My Zod validation is slow on large payloads. Switch to something faster."
The agent translates each Zod schema (z.string().email() becomes "string.email", z.coerce.number() becomes "string.numeric.parse" where the input is a string), swaps safeParse for the instanceof type.errors check, and times both libraries on a real payload before claiming a speedup.
Guidelines
- TypeScript syntax —
"string.email" not z.string().email()
type.errors for error checking — result instanceof type.errors.infer for TypeScript type — no duplicate interface definitions- Morphs for transforms —
.pipe() to transform during validation
- Scopes for complex schemas — define interconnected types with forward references
- Faster than Zod in the project's benchmarks — matters for hot paths; measure your own payloads
- Error messages are human-readable —
result.summary for display
? suffix for optional — "bio?": "string" makes bio optional- Constraints in the type —
"0 <= number < 100" is a range
string.numeric does not convert — use string.numeric.parse when you need a number- Cyclic types need an escape — make the back-reference optional or an array, or no finite value validates
- Not as battle-tested as Zod — newer library, smaller ecosystem