TypKit is a TypeScript validation library. You describe a schema by chaining readable constraints (or with an equivalent config object); the first time you run it, the whole tree is compiled into a single monomorphic function via new Function. Chaining is build-time only — it compiles to the identical validator — so the happy path allocates nothing, and rich constraints (maxLines, charset, blockWords, formats, ranges) compile to inline code instead of closures. Every schema is a Standard Schema and exports JSON Schema.

import { t } from '@myrialabs/typkit';

const Comment = t.object({
  author: t.string().minLength(3).maxLength(20).charset('a-z0-9_'),
  email: t.string().email(),
  body: t.string().nonEmpty().maxLines(500).blockWords(BANNED),
  rating: t.int().min(1).max(5),
  tags: t.string().array().optional(),
  visibility: t.enum(['public', 'private']),
});

Comment.check(value);                  // boolean — zero-allocation hot path
Comment.assert(value);                 // throws on failure, returns typed value
Comment.parse(value);                  // { value } | { issues }
Comment['~standard'].validate(value);  // Standard Schema — Elysia, Hono, tRPC
Comment.toJsonSchema();                // OpenAPI / JSON Schema
type Comment = t.infer<typeof Comment>;

bun add @myrialabs/typkit      # or: npm install @myrialabs/typkit

Why TypKit

• Rich constraints that stay fast. maxLines, charset, blockWords, allowedWords, and inline format checks (email, uuid, ipv4, Luhn creditCard, …) compile to literal loops. Competitors fall back to .refine()/.check()/.narrow() closures — 2.3–3.2× slower — and TypeBox can't express them at all.
• Zero-allocation hot path. check() returns a boolean; a validated value is never wrapped in a new object, and an issue path is built only when something fails. It sits in the compiled fast tier with ArkType and TypeBox, and runs 8–90× faster than Zod, Valibot, and effect/Schema — see bench-results.md.
• Works everywhere, no adapter. Standard Schema means Elysia, Hono, tRPC, react-hook-form, and TanStack Form accept a schema directly.
• JSON Schema in, OpenAPI out. toJsonSchema() maps 1:1 from the config object, so frameworks generate OpenAPI / Swagger straight from your validation schema.
• Honest type inference. t.infer<> is full and precise, and stays light on tsc (about a quarter of ArkType's instantiations).
• Self-documenting fluent API. Chain readable constraints — t.int().min(1).max(5) — with no cryptic gte/lt: min/max are inclusive, greaterThan/lessThan exclusive. Chaining is build-time only, so it compiles to the same validator as the equivalent config object and costs nothing at runtime.
• Zero runtime dependencies. Node 18+, Bun, and the browser.

Documentation

• Guide — from first schema to framework integration.
• API reference — every builder and method.
• Constraints — every inline rule and format.
• Frameworks — Elysia, Hono, tRPC, forms.
• JSON Schema — OpenAPI / Swagger export.
• Comparison — TypKit vs Zod, ArkType, TypeBox, Valibot.
• Benchmarks — reproducible numbers.

License

MIT © Myria Labs