Response

Every loader, .ctx, and middleware gets a set helper — the one surface for shaping the HTTP response from inside your code. With it you set headers, the status code, and cookies; the framework collects them and applies them to the final response. The simplest path doesn't even need set: return [statusCode, data] from a loader and the status is yours.

export const loginMutation = root.lets
  .mutation()
  .input(z.object({ email: z.string(), password: z.string() }))
  .loader(async ({ input, set }) => {
    const { user, token } = await auth.login(input)
    set.cookies('session', token, {
      httpOnly: true,
      secure: true,
      maxAge: 60 * 60 * 24,
    })
    set.headers('X-User-Id', user.id) // a response header
    set.status(201) // the HTTP status
    return { user } // becomes the mutation's data
  })
  .mutation()

set is available wherever a request runs server-side — loaders, .ctx, and middlewares all receive it. It does nothing useful on the client (the response is already sent); these writes only matter on the server, under SSR or for an endpoint call.

set.status — the HTTP status

.loader(({ set }) => {
  set.status(201) // any number; no validation, no clamping
  return { ok: true }
})

The status you write wins over the response's default 200. There's no shortcut for the common case, though — when the status tracks the data, return a tuple instead.

Return [statusCode, data]

A loader can return [status, data] and skip set.status entirely:

export const ideaPage = root.lets
  .page('/ideas/:id')
  .loader(() => [201, { x: 1 }]) // status 201, data { x: 1 }
  .page(({ data }) => <p>{data.x}</p>) // data is { x: number } — the status is stripped

The status sets the response code; the second element is the data your component or caller sees. The tuple's status is erased from the data type, so data stays { x: number }, never [number, …]. It works the same on a mutation:

.loader(() => [201, { x: 1 }]).mutation()
// result.response?.status === 201, result.data?.x === 1

If the data slot is undefined or null, the data normalizes to {}:

.loader(() => [204, undefined]) // status 204, data {}

set.headers — response headers

Three call forms, one method:

set.headers('X-User-Id', user.id) // name + value
set.headers({ 'Content-Type': 'text/plain', 'X-Custom': '1' }) // an object
set.headers(someHeadersInstance) // copy a whole Headers object

Two things to know:

set.headers('X-User-Id', '42')
set.inspect.headers['x-user-id'] // => '42'  — names are LOWERCASED
set.inspect.headers['X-User-Id'] // => undefined

set.headers('x-foo', undefined) // an undefined value DELETES the header from the response

A real use is a CORS middleware writing the allow-origin and vary headers:

set.headers('Access-Control-Allow-Origin', allowOriginValue)
set.headers('Vary', '*')

set.cookies — response cookies

Set a cookie by name and value, or as one options object:

set.cookies('session', token, {
  httpOnly: true,
  secure: true,
  maxAge: 60 * 60 * 24,
})
set.cookies({ name: 'theme', value: 'dark', sameSite: 'strict' })

Two defaults are filled in for you on every write:

set.cookies('session', 'abc123')
set.inspect.cookies.session
// => { name: 'session', value: 'abc123', path: '/', sameSite: 'lax' }

path defaults to '/' and sameSite to 'lax'. To let the browser pick the path, pass path: ''. Everything else (domain, secure, httpOnly, partitioned, maxAge, expires) passes through as given. When serialized into the Set-Cookie header, maxAge is floored to an integer, attribute values are truncated at the first ; (no injection), and an invalid sameSite falls back to 'lax'.

Delete a cookie by passing undefined as the value — both call forms work, and the framework expires it for you (Max-Age=0, Expires in the past):

set.cookies('session', undefined) // delete the cookie
set.cookies({ name: 'token', value: undefined }) // same, object form

set.cookies(...) is the low-level write. For a typed cookie with transformers, fallbacks, and a read/write API, use the CookieStore — it funnels its server writes through this same set.cookies under the hood.

Read request cookies — the ones the browser sent — from request.cookies['name'], not from set. set is for the response only. See Request.

set.inspect — read what's accumulated

set.inspect returns a fresh snapshot of the headers, cookies, and status set so far:

set.inspect // => { headers: {...}, cookies: {...}, status: 201 | undefined }

Each access is a new object with copied contents — reading it never mutates state, and header keys are lowercased here too. Because every middleware, .ctx, and loader on one request share the same effects, a later step reads what an earlier one wrote:

// a middleware sets a header...
.middleware(async ({ set, next }) => {
  set.headers('y', '3')
  return await next()
})
// ...and a page loader downstream reads it back
.loader(({ set }) => ({ x: 1, y: set.inspect.headers.y, z: set.inspect.headers.z }))
// renders x=1, y='3', z=undefined  (z was never set)

How effects accumulate

One effects collector lives per request. Every middleware, .ctx, and loader in the chain writes into it, so effects accumulate across the whole point execution — last write wins per key (headers by lowercased name, cookies by name, status is a single slot). At the very end, the framework applies the collected effects to the response. Because that apply runs after the whole chain has finished, a middleware that writes an effect after await next() returns still lands in the final response — and, being the last write for its key, it wins.

The framework writes effects too: a request-id header, input-validation 422, a 404 for an unknown point, a 500 for a point with no server loader, and (in dev only) an X-Point0-Renders-Count header — all land in the same collector.

Returning a Response (actions and mutations only)

For full control, return a native Response from a loader. This is allowed only on a mutation or an action — every other point type (page, layout, component, query, …) rejects a Response return at compile time:

export const apiHealthAction = root.lets
  .action('GET', '/api/health')
  .action(async () => {
    return new Response('OK', {
      status: 200,
      headers: { 'content-type': 'text/plain; charset=utf-8' },
    })
  })

// on a page, this is a type error:
root.lets.page('/x').loader(() => new Response('zxc'))
// Output can not be type of "Response" for point of type "page"

When a loader returns a Response, that response carries the result and the point has no data. Your set effects still apply to it — headers, cookies, and status set via set merge into the Response you returned. The same holds for a middleware: returning a Response short-circuits the chain, but effects set by earlier middlewares still land:

.middleware(async ({ set, next }) => { set.headers('y', '3'); return await next() })
.middleware(() => new Response('custom response'))
// final response: body 'custom response', status 200, header y === '3'

A .ctx cannot return a Response — only data-shaping, a redirect, or an error. Use a loader for that.

Who wins when both set something

When effects meet a returned Response, the Response's own values win:

• Headers — the Response's headers override effects headers of the same name.
• Cookies — a Set-Cookie on the Response keeps its cookie; an effects cookie of the same name is dropped, others append.
• Status — effects.status is used only if the Response didn't set one.

Error status under SSR

During server-side rendering, an error's HTTP status flows through the same effects collector, so the page responds with the right code.

A loader (or .ctx) throws or returns an error with a status:

.loader(() => { throw new AppError('gone', { status: 410 }) })
// SSR response.status === 410   (returning the error works the same)

AppError here is either the built-in ErrorPoint0 or your own error class — any class of the same-or-wider structure, wired in with .errorClass(...). A status on the error becomes the response status. (error0 is one optional way to build such a class, but nothing forces it.)

An error rendered by the error component sets the status during the render pass. Call setStatus from anywhere in a component — it's exported from @point0/core:

import { setStatus, useSetStatus } from '@point0/core'

setStatus(404) // safe to call anywhere — sets the status under SSR, no-op on the client

setStatus is safe to call on the client — it doesn't throw and it doesn't break anything. It only takes effect under SSR; on the client the status was already sent, so the call is a no-op. That's why you can call it straight from a component's render without guarding the side:

function NotFound() {
  useSetStatus(404) // identical to setStatus(404), just the hook-shaped name
  return <p>Not found</p>
}

useSetStatus is the same function under a use* name. Calling setStatus at the top of a component looks like a side effect during render, which trips React's rules-of-hooks lint; the use* alias quiets that lint so you can call it inline without warnings. It is not a real hook — same function, no extra behavior.

The framework also writes SSR status directly in a few spots: 404 when no page matches the URL, 422 on input-validation failure, and a redirect honors only 301 / 302 / 303 / 307 / 308 (anything else falls back to 302).

In production SSR the error stack is never rendered (only message/status). Error serialization is on Error handling.

The Effects object

set is the write surface, but it's only part of the per-request effects collector. That collector is an Effects instance, and you can reach it directly when you build a helper that has no set in scope (a plain server function):

import { getEffects, getEffectsOrUndefined } from '@point0/core'

const effects = getEffects() // the same collector every loader/ctx/middleware shares

Both are exported from @point0/core. getEffects throws when there's no request in scope; getEffectsOrUndefined returns undefined — that's exactly how setStatus stays a safe no-op off-request:

getEffects().set.status(204) // throws if called outside a request
getEffectsOrUndefined()?.set.headers('x-trace', id) // undefined outside a request

An Effects instance exposes more than just set:

So set is the curated way to write; effects.headers / effects.cookies / effects.status are the raw fields you can read directly, and effects.values (or set.inspect) is the safe immutable snapshot.

Reference

set surface

The object passed as set to loaders, .ctx, and middlewares:

set is a forbidden key for .ctx(..., { expose: [...] }) — you can't re-export it from ctx.

Cookie options

set.cookies accepts these (only name and value are required; value: undefined deletes):

A bare number for expires is epoch milliseconds — the same rule on the server (set.cookies) and in the browser (CookieStore). Use maxAge (seconds) for a relative lifetime.