Laravel Ai Tasks

Orchestrator AI-tasks for Laravel: drivers, routing, webhooks, queue, ai_runs, post-processing.

Installation

Install the package via composer:

composer require fomvasss/laravel-ai-tasks

Publish and run the migrations with:

php artisan vendor:publish --provider="Fomvasss\AiTasks\AiServiceProvider" --tag=config
php artisan vendor:publish --provider="Fomvasss\AiTasks\AiServiceProvider" --tag=migrations
php artisan migrate

Example .env:

OPENAI_API_KEY=...
GEMINI_API_KEY=...

Supervisor/Queue configuration

Add the following queues to your queue worker command (or use Horizon as below):

php artisan queue:work --queue=ai:webhook,ai:high,ai:default,ai:low,ai:post,ai:webhook

Horizon configuration

If you are using Laravel Horizon, you can configure the queues in config/horizon.php:

'environments' => [
    '*' => [
        // Prioryty online dialogues/streams (min. latency)
        'supervisor-ai-high' => [
            'connection'   => 'redis',
            'queue'        => ['ai:high'],
            'balance'      => 'auto',     // auto|simple|false
            'minProcesses' => 2,
            'maxProcesses' => 24,
            'tries'        => 1,
            'timeout'      => 40,         // seconds
            'nice'         => 0,
        ],

        // Typical text generations
        'supervisor-ai-default' => [
            'connection'   => 'redis',
            'queue'        => ['ai:default'],
            'balance'      => 'auto',
            'minProcesses' => 2,
            'maxProcesses' => 32,
            'tries'        => 3,
            'timeout'      => 120,
            'nice'         => 5,
        ],

        // Mass/slow tasks (catalogs, images)
        'supervisor-ai-low' => [
            'connection'   => 'redis',
            'queue'        => ['ai:low'],
            'balance'      => 'simple',
            'minProcesses' => 1,
            'maxProcesses' => 24,
            'tries'        => 2,
            'timeout'      => 300,
            'nice'         => 10,
        ],

        // Postprocessing tasks (validation JSON, conversions, saving)
        'supervisor-ai-post' => [
            'connection'   => 'redis',
            'queue'        => ['ai:post'],
            'balance'      => 'simple',
            'minProcesses' => 1,
            'maxProcesses' => 12,
            'tries'        => 2,
            'timeout'      => 300,
            'nice'         => 5,
        ],

        // Webhook from providers (short, reactive)
        'supervisor-ai-webhook' => [
            'connection'   => 'redis',
            'queue'        => ['ai:webhook'],
            'balance'      => 'simple',
            'minProcesses' => 1,
            'maxProcesses' => 8,
            'tries'        => 1,
            'timeout'      => 30,
        ],
    ],
]

Usage

Make Task

Use command to create new task:

php artisan ai:make-task SomeInterestingTask --modality=text

This will create new task class in app/Ai/Tasks/SomeInterestingTask.php.

Configure Task

<?php

namespace App\Ai\Tasks;

use Fomvasss\AiTasks\Contracts\QueueSerializableAi;
use Fomvasss\AiTasks\Contracts\ShouldQueueAi;
use Fomvasss\AiTasks\Tasks\AiTask;
use Fomvasss\AiTasks\DTO\AiPayload;
use Fomvasss\AiTasks\DTO\AiResponse;
use Fomvasss\AiTasks\Support\Prompt;
use Fomvasss\AiTasks\Support\Schema;

class SomeInterestingTask extends AiTask 
{
    public function name(): string
    {
        return 'some_interesting';
    }

    public function modality(): string
    {
        return 'text'; // text|chat|image|vision|embed
    }

    public function toPayload(): AiPayload
    {
        // TODO add your payload generation logic here        
        return new AiPayload(
            modality: $this->modality(),
            messages: [['role' => 'system', 'content' => 'You are a web programmer\'s assistant.'], [ 'role' => 'user', 'content' => 'Tell me something interesting.']],
            options:  ['temperature' => 0.3, 'model' => 'gpt-4o'], // model options
        );
    }

    public function postprocess(AiResponse $resp): array|AiResponse
    {
        // TODO add your post-processing logic here
        // Post-processing of responses (can be stored in a database/storage or other your own mechanism)
        // If you expect JSON — parse it and return an array
        return $resp;
    }
    
}

Run Task

<?php

use Fomvasss\AiTasks\Facades\AI;

// 1) Sync
$result = AI::send(new \App\Ai\Tasks\SomeInterestingTask());

// 2) Async
AI::queue(new \App\Ai\Tasks\SomeInterestingTask(), drivers: 'openai');

// 3) Direct driver usage
$payload = new \Fomvasss\AiTasks\DTO\AiPayload(
    modality: 'text',
    messages: [[ 'role'=>'user','content'=> 'Tell me something interesting' ]],
    options: ['temperature' => 0.3],
);
$context = new \Fomvasss\AiTasks\DTO\AiContext(
    tenantId:  '123456',
    taskName: 'interesting_task'
);
$result = app(\Fomvasss\AiTasks\Core\AiManager::class)->driver('gemini')
    ->send($payload, $context);

To perform async tasks and process webhooks, queues with names as specified in the configuration file must be launched ai.php section queues.

Commands

The package provides several Artisan commands to manage AI tasks:

• ai:budget — show tenant budget vs spent
• ai:runs — list recent ai_runs
• ai:retry — demo retry of failed runs
• ai:make-task — generate a new Ai task class
• ai:request — ad-hoc AI request (sync or queued)

Changelog

Please see CHANGELOG for more information on what has changed recently.

Credits

• fomvasss

License

The MIT License (MIT). Please see License File for more information.