Cogent

Cogent helps your Laravel app render models fast, without writing with() and withCount() everywhere.

You declare what a resource needs once — right on the resource — and Cogent loads it for every model, in one batched pass, just before it renders.

#[Includes(['author' => UserResource::class])]
#[Counts('comments')]
final class PostResource extends JsonResource
{
    use PreparesResources;

    public function toArray(Request $request): array
    {
        return [
            'id' => $this->id,
            'author' => new UserResource($this->whenLoaded('author')),
            'comments_count' => $this->whenCounted('comments'),
        ];
    }
}

// One query for posts, one for authors, one for comment counts — no matter how many posts.
return PostResource::collection(Post::query()->paginate());

Cogent works across the whole path of a Laravel object:

Area	What Cogent does
Eloquent / Fluent	Plug, include and count relations on collections already in memory — no extra queries.
Request (Input/Output)	Load only the relations a request asks for (?include=); prepare resource relations and counts before rendering; cache rendered output at the resource or controller level.
Event / Broadcast	Send models over WebSockets as their resources; restore queued models without re-querying; batch broadcasts and check payload size.

Installation

composer require webong/cogent

The service provider is auto-discovered. Requires PHP 8.3+ and Laravel 12 or 13.

---

Eloquent / Fluent — relations on collections already in memory

Cogent starts with collections you already have. Instead of letting Eloquent re-query relations, you attach, include, or count what's already loaded.

Plug relations you already resolved

$conversations->plug('channel', $channels->get($conversation->channel_id));

// Or resolve per model:
$conversations->plug('channel', fn (Conversation $c) =>
    $channels->get($c->channel_id));

// Only attach to models that don't have it loaded yet:
$conversations->plugMissing('channel', $channels);

Include related models already loaded

include() walks relations that are in memory and returns every related model, deduplicated so the same row is never held twice:

$graph = $conversations->include('channel', 'participants'); // no queries

// Fetch anything not loaded yet, batched by model class:
$conversations->includeMissing('channel.contacts');

Count relations in one query

batchCount() writes {relation}_count on every model with one grouped query:

$conversations->batchCount('messages');
// now $conversation->messages_count is available

---

Request (Input/Output) — prepare, cache, guard

Requested relations (?include=)

#[Includable] declares relations that load only when the request names them:

#[Includable(['author' => UserResource::class], 'comments')]
final class PostResource extends JsonResource
{
    use PreparesResources;
}

A client can then request ?include=author,comments.post. Anything not declared is ignored, so a client can never load more than the resource allows.

Need a different query parameter than ?include=? Register it once in a service provider:

Cogent::includeParameter('with');

The Cogent approach: unified query interface

Cogent provides one interface for JSON:API-style query params (filter[], sort, include), OpenAPI docs, and AsyncAPI docs — all driven by resource attribute declarations. You don't use spatie/laravel-query-builder directly; Cogent wraps it.

Declare what your resource accepts right on the resource class:

use Webong\Cogent\Attributes\Filterable;
use Webong\Cogent\Attributes\Includable;
use Webong\Cogent\Attributes\Sortable;
use Webong\Cogent\Concerns\PreparesResources;

#[Filterable(['title' => 'exact', 'body' => 'partial'])]
#[Sortable(['title' => 'asc', 'created_at' => 'desc'])]
#[Includable(['author' => UserResource::class], 'comments')]
final class PostResource extends JsonResource
{
    use PreparesResources;
}

Then query through Cogent::query() instead of spatie's QueryBuilder::for():

use Webong\Cogent\Cogent;

// GET /posts?filter[title]=Hello&sort=-created_at&include=author
$posts = Cogent::query(Post::query(), PostResource::class)->get();

// Or paginated:
$posts = Cogent::query(Post::query(), PostResource::class)->paginate(15);

// Or cursor-paginated:
$posts = Cogent::query(Post::query(), PostResource::class)->cursorPaginate(15);

// Access the underlying spatie QueryBuilder for advanced customization:
$spatie = Cogent::query(Post::query(), PostResource::class)->spatie();

Cogent::query() reads #[Filterable], #[Sortable], and #[Includable] from the resource class and applies them automatically through spatie. When the resource resolves, PreparesResources loads anything still missing and applies #[Counts] — spatie eager-loads the includes, Cogent fills in the gaps.

Attribute options:

Attribute	What it declares	Query param
#[Filterable(['title' => 'exact'])]	Filterable fields and types	filter[title]
#[Sortable(['title' => 'asc'])]	Sortable fields and directions	sort
#[Includable(['author'], 'comments')]	Includable relations	?include=

Filter types: exact, partial, beginsWith, endsWith, belongsTo, scope. Default: exact. Sort directions: asc, desc. Default: asc.

Read metadata for documentation

use Webong\Cogent\QueryBuilder\CogentQueryBuilder;

$metadata = CogentQueryBuilder::metadataFor(PostResource::class);

// ['author' => null, 'comments' => null]
$metadata['includes'];

// ['title' => 'exact', 'body' => 'partial']
$metadata['filters'];

// ['title' => 'asc', 'created_at' => 'desc']
$metadata['sorts'];

Pass it to PlanExporter::openApi() so OpenAPI output includes filter[] and sort parameters alongside ?include=:

use Webong\Cogent\Resources\PlanExporter;

$exporter = new PlanExporter();
$document = $exporter->openApi(
    [PostResource::class],
    'include',
    [PostResource::class => CogentQueryBuilder::metadataFor(PostResource::class)],
);

The resulting OpenAPI document gets filter[title], filter[body], and sort query parameters in components.parameters, plus an x-cogent extension carrying the filter type and sort direction for each.

Combined with #[Includable] on your resources, this gives you full JSON:API-style query support: filtering, sorting, and includes — all declared, all documented, all validated.

Declaring what a resource renders

Add PreparesResources to a resource and declare what it renders:

#[Includes(['author' => UserResource::class, 'channel'])]
#[Counts('comments')]
final class PostResource extends JsonResource
{
    use PreparesResources;

    public function toArray(Request $request): array
    {
        return [
            'id' => $this->id,
            'author' => new UserResource($this->whenLoaded('author')),
            'channel' => $this->whenLoaded('channel'),
            'comments_count' => $this->whenCounted('comments'),
        ];
    }
}

• #[Includes] loads the listed relations for every model, once, sharing duplicate instances.
• #[Counts] writes {relation}_count on every model with one grouped query.
• Relations already loaded are left alone.

Use it anywhere:

return PostResource::collection(Post::query()->paginate());
return Post::query()->get()->toResourceCollection(PostResource::class);
return new PostResource($post);

A single resource prepares its one model. Collections and paginators prepare every model in one batched pass.

Custom preparation with prepare()

For anything the attributes can't express, add a static prepare() hook. It runs once per resolve with every model:

public static function prepare(CogentCollection $posts, Request $request): void
{
    $posts->plugMissing('channel', $request->user()->channels);
}

Cache rendered output (#[Remember])

Cache what a resource renders so a cache hit skips preparing and toArray() entirely:

#[Remember(300)]
#[Includes(['author' => UserResource::class])]
#[Counts('comments')]
final class PostResource extends JsonResource
{
    use PreparesResources;
}

return PostResource::collection($posts);   // cached for 300 seconds
return PostResource::collection($posts)->remember(60); // or per call

Each authenticated user gets their own entry unless you mark it shared:

#[Remember(300, shared: true)]

Controller-level caching (Input / Output)

#[Remember] lives on the resource. If you want to cache an entire controller response — including anything the controller adds beyond the resource — use the remember() macro on the request or response, or inject OutputCache directly. This is an Input/Output cache: the input is the models plus anything the output varies on, and the output is the rendered array. It auto-forgets when any of those models (or rows reached through their loaded relations) is saved or deleted.

Configuration (config/cogent.php)

Cogent ships a config file with named cache surfaces, each with its own enabled flag and ttl:

return [
    'domains' => ['posts', 'users', 'comments', 'channels', 'conversations'],
    'registry' => [
        'posts' => ['list', 'counts', 'detail', 'scope'],
        'users' => ['list', 'detail', 'scope'],
        'comments' => ['list', 'counts'],
        'channels' => ['scope'],
        'conversations' => ['list', 'counts', 'context', 'scope', 'detail'],
    ],
    'surfaces' => [
        'list' => ['enabled' => false, 'ttl' => 30],
        'counts' => ['enabled' => false, 'ttl' => 20],
        'context' => ['enabled' => false, 'ttl' => 60],
        'scope' => ['enabled' => false, 'ttl' => 120],
        'detail' => ['enabled' => false, 'ttl' => 30],
    ],
];

Enable surfaces via environment variables:

COGENT_LIST_CACHE_ENABLED=true
COGENT_LIST_CACHE_TTL=30

Using the request macro (wraps the build closure)

public function index(Request $request)
{
    $posts = Post::query()->orderBy('created_at', 'desc')->get();

    return $request->remember(
        resource: PostResource::class,
        models: $posts->all(),
        vary: ['locale' => app()->getLocale(), 'page' => $request->input('page')],
        ttl: 300,
        build: fn () => PostResource::collection($posts)->resolve($request),
    );
}

Using the response macro (wraps an already-built response)

public function show(Post $post)
{
    return response()->json(new PostResource($post))->remember(
        resource: PostResource::class,
        models: [$post],
        ttl: 300,
    );
}

Injecting OutputCache directly

use Webong\Cogent\Resources\OutputCache;

public function index(OutputCache $cache, Request $request)
{
    $posts = Post::query()->orderBy('created_at', 'desc')->get();

    return $cache->remember(
        resource: PostResource::class,
        models: $posts->all(),
        vary: ['locale' => app()->getLocale(), 'page' => $request->input('page')],
        ttl: 300,
        build: fn () => PostResource::collection($posts)->resolve($request),
    );
}

Attribute-based caching on controller methods

Configure caching declaratively with #[RememberResponse] on a controller method. Register the remember.response middleware on the route:

use Webong\Cogent\Attributes\RememberResponse;

// In routes/web.php or routes/api.php:
Route::get('/posts', [PostController::class, 'index'])
    ->middleware('remember.response');

Then on the controller method:

use Webong\Cogent\Attributes\RememberResponse;

#[RememberResponse(surface: 'list', domain: 'posts', vary: ['locale', 'page'])]
public function index(Request $request)
{
    $posts = Post::query()->orderBy('created_at', 'desc')->get();

    return PostResource::collection($posts)->resolve($request);
}

Options:

Option	Description
surface	Named cache surface (list, counts, detail, scope, context). Must be enabled in config/cogent.php.
domain	Cache domain grouping related resources (posts, users, comments, etc.).
vary	Request input keys or closures that differentiate the cache entry.
shared	One entry for everyone instead of per authenticated user.
when	Closure receiving the request; only cache when it returns true.

vary accepts request input keys (strings) or closures that receive the request and return a value:

#[RememberResponse(
    surface: 'list',
    domain: 'posts',
    vary: [
        'locale',
        'page',
        fn (Request $request) => $request->header('x-app-version'),
    ],
)]

Only cache when a condition is met:

#[RememberResponse(
    surface: 'detail',
    domain: 'posts',
    when: fn (Request $request) => $request->query('cache') === '1',
)]

Domains and surfaces used in #[RememberResponse] attributes are automatically registered into config/cogent.php at runtime.

Make sure nothing lazy-loads while rendering

Turn on the dev guard so any query that sneaks in while a resource renders throws instead of silently becoming an N+1:

// In a service provider's boot():
Cogent::guardResources(! app()->isProduction());

Cogent::handleResourceQueriesUsing(function (string $resource, \Illuminate\Database\Events\QueryExecuted $query): void {
    logger()->warning("{$resource} queried while rendering", ['sql' => $query->sql]);
});

---

Event / Broadcast — WebSockets and queues

Broadcast models through their resources

Send models over WebSockets as their resources, so clients get the same shape over a socket as over HTTP:

use Webong\Cogent\Attributes\WithResource;
use Webong\Cogent\Concerns\BroadcastsResources;

final class PostUpdated implements ShouldBroadcast
{
    use BroadcastsResources;

    public function __construct(
        public Post $post,
        #[WithResource(EditorResource::class)] public User $editor,
    ) {}
}

Models without a resource fall back to toArray().

For models that broadcast their own events, use BroadcastsAsResource instead of Laravel's BroadcastsEvents — every model event is sent as the model's resource:

#[UseResource(PostResource::class)]
class Post extends Model
{
    use BroadcastsAsResource;
}

Batch many model broadcasts so every model is prepared in one pass and nothing is sent if the callback throws:

Cogent::batchBroadcasts(function () use ($posts): void {
    $posts->each->update(['published' => true]);
});

Check payloads against the driver limit (Pusher 10KB, Reverb ~10,000 bytes):

Cogent::limitBroadcasts(10_000);

Cogent::handleOversizedBroadcastsUsing(function (object $event, int $bytes, int $limit): void {
    logger()->warning($event::class." broadcast is {$bytes} bytes, over {$limit}");
});

Queued events and jobs

Laravel's SerializesModels re-queries every restored model, losing counts and cached relations. Use SerializesPreparedModels instead — restored models load what their resource renders, and cacheable relations come from the cache:

use Webong\Cogent\Concerns\SerializesPreparedModels;

final class PostUpdated implements ShouldBroadcast
{
    use BroadcastsResources;
    use SerializesPreparedModels;

    public function __construct(
        public Post $post,
        #[WithResource(EditorResource::class)] public User $editor,
    ) {}
}

Broadcast documentation (AsyncAPI)

AsyncAPI is documentation only — it describes what your broadcast events look like; it doesn't handle WebSocket connections at runtime (that's Reverb/Pusher's job).

Generate an AsyncAPI 3.0 document so frontend teams know what channels to subscribe to and what payload shapes to expect:

php artisan cogent:plans --format=asyncapi \
    --events-path=app/Events \
    --models-path=app/Models \
    --title="My Application WebSockets"

It scans events using BroadcastsResources/SnapshotsBroadcasts and models using BroadcastsAsResource, producing:

• Channels — where events broadcast (public, private, presence), with dynamic addresses marked
• Operations — send actions for each event
• Messages — event names and payload schemas
• Schemas — the resource shapes broadcast over the socket

The OpenAPI and AsyncAPI formats require composer require --dev dedoc/scramble.

---

Artisan commands

# Human-readable table of what every preparing resource loads, counts and caches
php artisan cogent:plans

# Same as JSON, for tooling
php artisan cogent:plans --format=json

# OpenAPI 3.1 components: include parameters, plans and Scramble-inferred response schemas
php artisan cogent:plans --format=openapi

# AsyncAPI 3.0 document: broadcast channels, event names and resource payloads
php artisan cogent:plans --format=asyncapi \
    --events-path=app/Events --models-path=app/Models --title="My broadcasts"

Scan custom directories:

php artisan cogent:plans --path=app/Http/Resources/Api --format=json
php artisan cogent:plans --format=asyncapi --events-path=app/Events --models-path=app/Models

The OpenAPI and AsyncAPI formats require composer require --dev dedoc/scramble.

Running the tests

composer install
composer test

License

MIT.