Clean API Client
The main rule of reliable architecture: the request layer must be an independent module. It should know nothing about components, pages, or UI frameworks.
The Problem: Chaos in Components
When requests are made directly inside components, the project accumulates technical debt very quickly. You'll inevitably run into these problems:
- Duplication: the same endpoint is called across multiple different files.
- Inconsistency: error handling and header configuration differ everywhere.
- Painful refactoring: when the API changes, you have to rewrite half the UI components.
- Poor reusability: the logic can't be moved to another project.
I avoid this with a simple, flat structure that strictly separates transport, domains, and the entry point.
Request Layer Architecture
Basic Directory Structure
import * as products from "@/services/requests/products/index"
import * as users from "@/services/requests/users/index"
export const ApiClient = {
products,
users
}
Each element in this structure has exactly one responsibility:
| File | Responsibility |
|---|---|
request.ts | Low-level transport: fetch/axios, timeouts, error normalization, and headers. |
requests/<domain> | Domain logic: endpoints (routes.ts) and typed methods (index.ts) for a specific entity. |
client.ts | A single facade (ApiClient) that assembles all domains into one convenient entry point. |
Request Layer (Transport)
request.ts contains a universal wrapper that assembles URLs with query parameters, sets timeouts, and parses responses.
Example implementation:
export const request = {
get: <T>(url: string, options?: Omit<ApiRequestOptions, "method">) =>
apiRequest<T>(url, { ...options, method: "GET" }),
post: <T>(url: string, body?: unknown, options?: Omit<ApiRequestOptions, "method" | "body">) =>
apiRequest<T>(url, { ...options, method: "POST", body }),
put: <T>(url: string, body?: unknown, options?: Omit<ApiRequestOptions, "method" | "body">) =>
apiRequest<T>(url, { ...options, method: "PUT", body })
}
Domain Modules
Each data source lives in its own folder. For example, products reside in products/. We separate the requests themselves from their paths so that URLs don't spread throughout the code.
export const routes = {
list: () => `/products`,
byId: (id: number) => `/products/${id}`
}
In index.ts we use these routes and our base transport, wrapping everything in strict types:
import { request } from "@/services/request"
import { routes } from "./routes"
import type { Product } from "./types"
export const getProducts = async (): Promise<Product[]> => {
try {
return await request.get<Product[]>(routes.list())
} catch (error) {
// Centralized handling or fallback
return []
}
}
export const getProduct = async (id: number): Promise<Product | null> => {
try {
return await request.get<Product>(routes.byId(id))
} catch (error) {
return null
}
}
Unified API Client
When there are many modules, importing each one individually becomes inconvenient. I assemble them into a shared client:
import * as products from "@/services/requests/products/index"
import * as users from "@/services/requests/users/index"
export const ApiClient = {
products,
users
}
Now the application has a predictable and convenient data access interface: ApiClient.<domain>.<method>.
How It's Used in the UI
See how much cleaner the component code becomes. It no longer knows anything about URLs, headers, tokens, or fetch internals.
<script setup lang="ts">
import { ref, onMounted } from "vue"
import { ApiClient } from "@/services/client"
const products = ref([])
const loading = ref(true)
onMounted(async () => {
try {
products.value = await ApiClient.products.getProducts()
} finally {
loading.value = false
}
})
</script>
The component only deals with its direct responsibility — managing state and rendering data.
Conclusion
This separation prevents mixing business logic with presentation. The basic set of a shared request, domain folders, and a single ApiClient is more than enough to keep the code from spreading and remain readable even as the project scales significantly.