REST API design tips

Best practices and tips for designing RESTful APIs, including naming conventions, versioning, and endpoint structuring.

Tips

Always have separate delete endpoint to delete a resource even for soft delete where you just set is_deleted flag to true.
Never allow PATCH to delete resource.
Send 204 to already deleted resource without raising exception of resource already deleted ( Idempotency principles).

Naming Conventions

Use kebab-case for URL paths and snake_case or camelCase for query parameters.

When using hyphens in query params, it can clash with minus‑sign semantics in some contexts. most libraries won't parse them as identifiers. Hence, avoid hyphens in query keys.

POST /sign-in
POST /sign-in?redirectUrl=https://example.com/dashboard # camelCase query params when JS frameworks
POST /sign_in?redirect_url=https://example.com/dashboard # snake_case query params when Python, Ruby, etc.

2. Query Parameters

You have two main, equally valid options—pick one and be consistent:

A. snake_case

Lowercase + underscores
Common in many REST APIs and frameworks

Example:

GET /products?category=electronics&price_min=100&sort_by=popularity

B. camelCase

Lowercase first word, then capital letters for subsequent words
Aligns with many JavaScript front‑end conventions

Example:

GET /products?category=electronics&priceMin=100&sortBy=popularity

A Note on Hyphens in Query Strings

Hyphens can clash with minus‑sign semantics in some contexts; most libraries won't parse them as identifiers.
For that reason, avoid hyphens in query keys.

3. General Tips

Keep names semantic
- Use real words: sortBy, filter, pageSize, lang
Document your choice in your API style guide so every team follows the same format.
Versioning: if you version in the URL, do it as its own segment:
```
/v1/products
```

API endpoint should be resource not action

https://api.example.com/tasks // [!code ++]

https://api.example.com/get-tasks // [!code --]

Collection name should be plural

https://api.example.com/tasks // [!code ++]
https://api.example.com/tasks/{task_id} // [!code ++]

https://api.example.com/task // [!code --]
https://api.example.com/task/{task_id} // [!code --]

POST /auth/register
POST /auth/login
POST /auth/verification-emails
GET /auth/verification-emails/{token}
POST /auth/forgot-password
POST /auth/reset-password

Entity CRUD endpoints conventions

GET    /tasks           # Get all tasks
POST   /tasks           # Create new task
GET    /tasks/{task_id} # Get task by id
PUT    /tasks/{task_id} # Update task by id (Replace)
PATCH  /tasks/{task_id} # Partial update task by id
DELETE /tasks/{task_id} # Delete task by id

POST   /tasks/batch     # Create multiple tasks

GET    /tasks/{task_id}/comments  # Get all comments for task
POST   /tasks/{task_id}/comments  # Create new comment for task
GET    /comments/{comment_id}     # Get comment by id for task
PUT    /comments/{comment_id}     # Update comment by id for task
PATCH  /comments/{comment_id}     # Partial update comment by id for task
DELETE /comments/{comment_id}     # Delete comment by id for task

Don't go deeper than 3 levels: `collection/resource/collection`

https://api.example.com/tasks/1/comments // [!code ++]
https://api.example.com/tasks/1/comments/1/replies // [!code --]

Filter, Sorting & Pagination

<!-- Filter: Use key-value pair -->
https://api.example.com/tasks?completed=true
https://api.example.com/users?last_name=john&age=36

<!-- Filter: Provide filtering on specific property/column/field with global search -->
https://api.example.com/users?q=john
https://api.example.com/tasks?q=john&fields=username,email

<!--
    Pagination: Prefer using limit-offset instead of page-size

    ?page=1&size=10 => fine for simple pagination
    🤔 => How to get 10 items starting from 4th item with page-size?
-->
https://api.example.com/tasks?limit=10&offset=0
https://api.example.com/tasks?limit=10&offset=3 <- You can get 10 items starting from 4th item with limit-offset

<!-- Sorting: sorting key with order -->
https://api.example.com/tasks?sort=-created_at,title

Instead of returning list return object with name for making your response future proof

# --- Before
[
    { "id": 1, "title": "Buy Tomato", "is_completed": False },
    { "id": 2, "title": "Learn Sanskrit", "is_completed": False },
]

# --- After
{
    "tasks": [
        { "id": 1, "title": "Buy Tomato", "is_completed": False },
        { "id": 2, "title": "Learn Sanskrit", "is_completed": False },
    ],
    "total": 2,
}

Don't return map structure in response, instead return list of objects

# --- Before
{
    "john": [
        { "id": 1, "title": "Buy Tomato", "is_completed": False },
        { "id": 2, "title": "Learn Sanskrit", "is_completed": False },
    ],
    "jane": [
        { "id": 3, "title": "Buy Apple", "is_completed": False },
        { "id": 4, "title": "Finish reading book", "is_completed": False },
    ]
}

# --- After
{
    "tasks": [
        { "id": 1, "title": "Buy Tomato", "is_completed": False, "user": "john" },
        { "id": 2, "title": "Learn Sanskrit", "is_completed": False, "user": "john" },
        { "id": 3, "title": "Buy Apple", "is_completed": False, "user": "jane" },
        { "id": 4, "title": "Finish reading book", "is_completed": False, "user": "jane" },
    ],
    "total": 4
}

Async/Longer Operations

For longer processes (POST { processId: 99 } => /long-process/create) Send 202 status code to indicate that request has been accepted for processing
- Add status endpoint URL in response header Location to check the status of the operation
Expose GET (/long-process/status/99) endpoint to check the status of the operation
- Return 200 status code if operation isn't completed with any of the below response:
  - Provide status of the operation
```
{ status: "In Progress" }
```
  - Provide estimated time to complete
```
{ "time_to_complete_in_minutes": "15" }
```
  - You can also provide link to cancel/stop the operation
```
{
    "rel": "cancel",
    "method": "delete",
    "href": "/long-process/cancel/99"
}
```
- Return 303 status code with Location header with URL to created resource once operation is completed
```
{ "status": "completed" }
```

API response should be generic and agnostic

Even if you are building API for a specific client, make sure your API response is generic and agnostic
Don't return client specific data in response. Doing so will make your API tightly coupled with the client and requires changing both client and server code whenever requirement changes.
E.g. For chat page, instead of single endpoint for chat details, chat messages, meta data, etc. create separate endpoints for each of them.

SEO

Hyphens improve keyword recognition in URL (excluding query parameters). E.g. /product-details

AI Engineering - My Findings

Learn useful tips and best practices in AI engineering.