README

A Laravel AI API package — connect to any LLM provider with a clean, unified interface.
Supports OpenAI, Claude, Gemini, Ollama, OpenRouter, Groq, DeepSeek, Mistral, Together AI, Fireworks, xAI, Cohere, Z.AI, and custom providers.

Requirements

• PHP >= 8.1
• Laravel >= 9.x

Installation

composer require mojahed/aiapi

Publish config:

php artisan aiapi:setup

Configuration

Add to your .env:

AIAPI_PROVIDER=openrouter
AIAPI_MODEL=z-ai/glm-5.2
# Note: "z-ai/glm-5.2" is the OpenRouter slug.
# For the native "zai" provider use the model name directly, e.g. glm-5.
AIAPI_BEHAVIOUR="You are a helpful assistant. Answer clearly and formally."

OPENROUTER_API_KEY=sk-or-...
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GEMINI_API_KEY=...
GROQ_API_KEY=...
DEEPSEEK_API_KEY=...
MISTRAL_API_KEY=...
TOGETHER_API_KEY=...
FIREWORKS_API_KEY=...
XAI_API_KEY=...
COHERE_API_KEY=...
ZAI_API_KEY=...

Basic Usage

use Mojahed\AIAPI\AI;

$ai = new AI();
$response = $ai->ask('What is Laravel?');

echo $response->answer;      // the answer
echo $response->summary;     // updated conversation summary
echo $response->tokensUsed;  // total tokens used
echo $response->model;       // model that responded
$response->raw;              // full raw API response array

Fluent Chaining

use Mojahed\AIAPI\AI;

$response = (new AI())
    ->setProvider('openrouter')
    ->setModel('z-ai/glm-5.2')
    ->setBehaviour('You are an Eduplus school assistant. Reply formally.')
    ->setKnowledge('/path/to/eduplus-docs.md')
    ->setSummary($summaryFromDB)
    ->setMaxTokens(1024)
    ->setTemperature(0.7)
    ->ask('How does the attendance system work?');

echo $response->answer;

Knowledge Injection

Pass a file path, raw string, or array of both:

// Single file
$ai->setKnowledge('/knowledge/eduplus.md');

// Raw string
$ai->setKnowledge('MdsChatBot is a floating chatbot widget...');

// Multiple files
$ai->setKnowledge([
    '/knowledge/git.md',
    '/knowledge/php.md',
    '/knowledge/eduplus.md',
]);

Summary Memory (Cost-Efficient)

Each response returns a compact summary. Pass it back on the next question:

// Load summary from DB
$summary = AiSession::find($sessionId)?->summary ?? '';

$response = (new AI())
    ->setSummary($summary)
    ->ask($question);

// Save updated summary to DB
AiSession::updateOrCreate(
    ['id' => $sessionId],
    ['summary' => $response->summary]
);

return $response->answer;

This gives the AI conversation memory at minimal token cost — no full history needed.

Switch Provider Per Instance

// Different providers for different purposes
$fast    = (new AI())->setProvider('groq');     // fastest
$smart   = (new AI())->setProvider('claude');   // most capable
$cheap   = (new AI())->setProvider('deepseek'); // cheapest
$local   = (new AI())->setProvider('ollama');   // free, local

Custom Provider

Implement ProviderInterface with 4 methods:

use Mojahed\AIAPI\Providers\ProviderInterface;

class MyCustomProvider implements ProviderInterface
{
    public function __construct(protected string $apiKey) {}

    public function send(array $messages, array $options): array
    {
        // Make your API call here
        // Return the raw response as array
    }

    public function extractContent(array $raw): string
    {
        return $raw['my_answer_field'] ?? '';
    }

    public function extractTokens(array $raw): int
    {
        return $raw['usage']['total'] ?? 0;
    }

    public function extractModel(array $raw): string
    {
        return $raw['model'] ?? 'custom';
    }
}

Register in AppServiceProvider::boot():

use Mojahed\AIAPI\AI;

public function boot(): void
{
    AI::registerProvider('myprovider', MyCustomProvider::class);
}

Use it:

$ai = (new AI())->setProvider('myprovider');

Test Connection

php artisan aiapi:test
php artisan aiapi:test --provider=claude
php artisan aiapi:test --provider=ollama --question="What is PHP?"

Supported Providers

Default Config (config/aiapi.php)

return [
    'provider'    => env('AIAPI_PROVIDER', 'openrouter'),
    'model'       => env('AIAPI_MODEL', 'z-ai/glm-5.2'),
    'behaviour'   => env('AIAPI_BEHAVIOUR', 'You are a helpful assistant.'),
    'knowledge'   => null,
    'max_tokens'  => 1024,
    'temperature' => 0.7,
    'timeout'     => 60,
    'keys'        => [
        'openrouter' => env('OPENROUTER_API_KEY'),
        // ...
    ],
];

Example: Full Chat Application

A complete, copy-paste-ready chat app lives in example/:

• example/migrations/ — chats + chat_messages tables
• example/Models/ — Chat & ChatMessage Eloquent models
• example/Http/Controllers/ChatController.php — chat with history + summary memory
• example/routes/chat.php — API routes

The optimization

Full message history is stored in the database (for the UI), but only the rolling summary is sent to the LLM on each turn. Token cost stays roughly flat per message instead of growing with the conversation length.

$ai = (new AI())
    ->setSummary($chat->summary)            // memory from previous turns (cheap)
    ->setBehaviour(config('aiapi.behaviour'));

$response = $ai->ask($message);

$chat->update(['summary' => $response->summary]);   // roll memory forward

Setup

1. Copy the files from example/ into the matching app folders.
2. Run php artisan migrate.
3. Register the routes from example/routes/chat.php inside routes/api.php.
4. Wrap the ask() call in a try/catch (see ChatController::ask) to show users a friendly error on rate limits / outages.

License

MIT