How Bagisto Builds Magic AI on Top of Laravel AI SDK

dhananjay.laravel311

19 June 2026

Bagisto introduces Magic AI, a built-in AI layer that enables features like product description generation, image creation, and visual search.

It also powers review translation and personalized checkout messages, bringing AI directly into the eCommerce workflow.

The Problem Magic AI Solves

Laravel AI provides the core SDK for interacting with AI models.

It doesn’t handle provider management, admin-configured credentials, or eCommerce-specific integrations.

Magic AI builds on top of Laravel AI to solve these challenges, providing a unified AI layer that works seamlessly across Bagisto.

Three Design Decisions

1. Every Model Is a Typed Enum

Instead of storing model names in config arrays, Bagisto represents every AI model as a PHP enum case implementing AiModelContract:

Each provider has its own enum containing all its models. Here’s how OpenAI implements it:

enum OpenAiModel: string implements AiModelContract

{

// Text models

case GPT52 = 'gpt-5.2';

case GPT41 = 'gpt-4.1';

// ... more text models

// Image models

case GptImage15 = 'gpt-image-1.5';

case GptImage1 = 'gpt-image-1';

public function provider(): Lab

{

return Lab::OpenAI;

}

public function isImageModel(): bool

{

return match ($this) {

self::GptImage15, self::GptImage1 => true,

default => false,

};

}

2. A Single-Array Provider Registry

The AiProvider class maps every provider to its model enum in one array.

AiProvider auto-generates everything the admin UI needs: provider dropdowns, model selectors (prefixed as “OpenAI: GPT-4.1”), and filtered lists.

interface AiModelContract

{

public function provider(): Lab; // Which SDK provider

public function label(): string; // Display name for admin UI

public function isImageModel(): bool; // Can generate images?

public function isTextModel(): bool; // Can generate text?

}

private static array $providers = [

['provider' => Lab::OpenAI, 'label' => 'OpenAI', 'model_enum' => OpenAiModel::class],

['provider' => Lab::Gemini, 'label' => 'Gemini', 'model_enum' => GeminiModel::class],

['provider' => Lab::Anthropic, 'label' => 'Anthropic', 'model_enum' => AnthropicModel::class],

// ...more entries, same pattern

];

3. Model-First Provider Resolution

Most AI wrappers ask: “Which provider? OK, now which model?”

Magic AI approaches the problem from the opposite direction. Instead of starting with a provider, the selected model determines which provider should be used.

Once resolved, the corresponding API credentials are loaded, and the request is passed to the SDK.

// Resolve the model string to its enum case

$modelEnum = AiProvider::resolveModel('gpt-4.1'); // → OpenAiModel::GPT41

$provider = $modelEnum->provider()->value; // → 'openai'

// Inject the admin-stored API key into Laravel AI's runtime config

$apiKey = core()->getConfigData("magic_ai.providers.{$provider}.api_key");

config(["ai.providers.{$provider}.key" => $apiKey]);

// Now call the SDK — it has the right credentials

agent()->prompt($prompt, provider: $provider, model: 'gpt-4.1')

Five eCommerce Features, One Pattern

Magic AI powers five AI-driven features across Bagisto, from content generation to visual search.

Despite their different use cases, they all follow the same pattern: build a prompt, select a provider, and execute the request through Laravel AI.

// Product & CMS Content Generation

agent()->prompt($contentPrompt);

// Product Image Generation

Image::of($imagePrompt);

// Search by Image

agent()

->withAttachment($uploadedImage)

->prompt($analysisPrompt);

// Review Translation

agent()->prompt("Translate to {$locale}: {$review}");

// Personalized Checkout Messages

agent()->prompt(buildCheckoutPrompt($order));

Before & After: AI-Generated Product Description

Without AI (manually written):

"Blue cotton t-shirt. Available in sizes S, M, L, XL. Machine washable."

With AI (generated via GPT-4.1 from admin panel):

"Elevate your everyday wardrobe with this classic blue cotton t-shirt.

Crafted from 100% breathable cotton, it offers all-day comfort with a relaxed fit that pairs effortlessly with jeans, chinos, or layered under a blazer.

Available in S through XL. Easy care — simply machine wash and tumble dry."

Same product, dramatically better copy — generated in seconds from the product editor.

Admin Configuration at a Glance

The admin panel exposes Magic AI settings in four sections, all defined declaratively in system.php:

The model dropdowns are populated dynamically by AiProvider::textModelOptions(), which aggregates all text-capable models across every registered provider.

Configuration → Magic AI

├── General → Master on/off toggle

├── Providers → API key fields for all 8 providers (+ Ollama URL)

├── Admin Features → Enable text/image generation + select which providers are available

└── Storefront → Enable image search, review translation, checkout messages

+ select which specific model powers each feature

Storefront Integration

The Shop package integrates Magic AI with simple feature checks. Both follow the same pattern: check the feature toggle before calling Magic AI.

// SearchController — Image Search

if (core()->getConfigData('magic_ai.storefront_features.image_search.enabled')) {

$keywords = magic_ai()->analyzeImage($uploadedImagePath);

// Feed keywords into product search...

}

// OnepageController — Checkout Message

if (core()->getConfigData('magic_ai.storefront_features.checkout_message.enabled')) {

$message = magic_ai()->checkoutMessage($order);

}

How to Add a New AI Provider

Adding a new provider requires two changes:

Step 1: Create a Model Enum

Create a new PHP enum inside Enums/Models/ implementing AiModelContract:

enum NewProviderModel: string implements AiModelContract

{

case ModelA = 'new-provider-model-a';

case ModelB = 'new-provider-model-b';

public function provider(): Lab { return Lab::NewProvider; }

public function label(): string { /* ... */ }

public function isImageModel(): bool { /* ... */ }

public function isTextModel(): bool { return !$this->isImageModel(); }

public static function defaultTextModel(): ?static { return self::ModelA; }

public static function defaultImageModel(): ?static { return null; }

}

Step 2: Register in AiProvider

Add one line to the $providers array: The admin UI auto-populates, the API key config field appears, and credential injection works automatically.

No controller changes, no route changes, no migration.

1	['provider' => Lab::NewProvider, 'label' => 'New Provider', 'model_enum' => NewProviderModel::class],

The Layer Architecture

Magic AI

Conclusion

Bagisto’s Magic AI demonstrates a clean pattern for building framework-native AI integrations:

Foundation — Laravel AI SDK provides agent() and Image::of()
Abstraction — Magic AI wraps them with eCommerce features and model-first provider resolution
Configuration — The admin panel gives non-developers a GUI to manage everything

The architecture is worth studying beyond eCommerce.

Laravel applications can use this enum registry and runtime injection pattern to support multiple AI providers without hard-coded credentials or vendor lock-in.

You can also explore our Bagisto Extensions. If you are planning to build with Laravel, consider hiring laravel developers .