upfetch

up-fetch

Mental model

• up(fetchFn, getDefaultOptions?)

  creates the reusable client.

• getDefaultOptions(input, options, ctx)

  runs on every request.

• upfetch(input, options?, ctx?)

  performs one request.
• Keep

  SKILL.md

  high-level; load the relevant file under

  references/

  for details.

Minimum pattern

import { up } from 'up-fetch'
import { z } from 'zod'

export const upfetch = up(fetch, () => ({
   baseUrl: 'https://api.example.com',
   headers: {
      Authorization: readToken() ? `Bearer ${readToken()}` : undefined,
   },
   timeout: 5000,
}))

const user = await upfetch('/users/1', {
   schema: z.object({
      id: z.number(),
      name: z.string(),
   }),
})

Workflow

1. Start with client setup and dynamic defaults.
2. If the request needs auth, params, body shaping, or merge semantics, read auth and request shaping.
3. If the response contract matters, read validation, parsing, and errors.
4. If retries, timeouts, or hook timing matter, read retries, timeouts, and lifecycle.
5. If streaming or runtime quirks matter, read streaming and runtime caveats.

High-value rules

• Pass a function as the second argument to

  up()

  , not a plain object.
• Read auth and other mutable defaults inside that function so values stay fresh.
• Use

  params

  and

  body

  instead of hand-serializing query strings or JSON.
• Use

  schema

  when you need runtime trust; TypeScript generics alone do not validate payloads.
• If you want error-as-value behavior, set

  reject: () => false

  before relying on

  parseResponse

  .
• Prefer

  globalThis.fetch

  over imported

  undici.fetch

  .

References

• Client setup and dynamic defaults
• Auth and request shaping
• Validation, parsing, and errors
• Retries, timeouts, and lifecycle
• Streaming and runtime caveats