symfony.md

Symfony integration

ferry-ai/inference-symfony provides a bundle and a DI extension that configure FerryAI in a Symfony application.

Like the Laravel adapter, the Symfony classes are decoupled: AIBundle does not extend Symfony's Bundle, Configuration does not implement ConfigurationInterface, and FerryAIExtension does not extend Extension — no hard Symfony dependency. Add symfony/http-kernel, symfony/config, symfony/dependency-injection yourself (suggest) to wire them natively.

Package

packages/symfony/src/
├── AIBundle.php                          # Bootstraps FerryAI from config
└── DependencyInjection/
    ├── Configuration.php                 # Defines the config tree
    └── FerryAIExtension.php              # Loads config and sets up container

Bundle

AIBundle::boot(array $configs = []) merges the given config over the bundle defaults and calls AI::config(). FerryAIExtension::load() performs the same wiring inside a Symfony container. Configuration::getConfigTree() returns the default tree shown below; any additional AIConfig key you supply (e.g. embedding.*, vector.*, model_pool.*, observability.*) is passed straight through to AI::config().

Configuration

Bundle defaults (from Configuration::getConfigTree()):

# config/packages/ferry_ai.yaml
ferry_ai:
    backend: auto
    device: auto
    model_cache: '%kernel.project_dir%/var/models'
    max_tokens: 2048
    temperature: 0.7
    top_p: 1.0
    verify_signatures: true
    backends:
        onnx:
            providers: ['CUDA', 'CPU']
            graph_optimization: ALL
        llama:
            model_path: ~
            n_ctx: 2048
            n_gpu_layers: 0
    warmup: []
    log_channel: stack

Additional keys accepted by AI::config() (not part of the default tree, but forwarded if present) — see configuration for the full list:

ferry_ai:
    backends:
        embedding: { model_path: '%kernel.project_dir%/models/all-MiniLM-L6-v2-onnx' }
        classify:  { model_path: ~ }
    embedding: { pooling: mean, normalize: true }
    vector:    { driver: sqlite, db_path: ':memory:', metric: cosine }
    observability: { metrics: false, profiling: false, logging: false }

Register the bundle

In config/bundles.php (or via manual registration):

return [
    // ...
    FerryAI\Symfony\AIBundle::class => ['all' => true],
];

Or manually:

$bundle = new FerryAI\Symfony\AIBundle();
$bundle->boot();   // reads config and calls AI::config()

Use in services

Once booted, use the facade anywhere:

<?php

namespace App\Controller;

use FerryAI\AI;
use Symfony\Component\HttpFoundation\JsonResponse;
use Symfony\Component\HttpFoundation\Request;

class InferenceController
{
    public function embed(Request $request): JsonResponse
    {
        $text = $request->get('text');
        $vec  = AI::embed($text);

        return new JsonResponse([
            'dimension' => $vec->dimension,
            'model'     => $vec->modelName,
        ]);
    }

    public function chat(Request $request): JsonResponse
    {
        $prompt = $request->get('prompt');
        $reply  = AI::chat([['role' => 'user', 'content' => $prompt]]);

        return new JsonResponse(['reply' => $reply->text]);
    }
}

Accessing FerryAI

The adapter does not register container services; use the static FerryAI\AI facade directly from any service or controller once the bundle has booted. To inject FerryAI behind your own service, create thin wrappers in your app and call AI::embed() / AI::vector() inside them:

// config/services.yaml
services:
    App\Service\RagService: ~   # calls FerryAI\AI internally

See examples/20-symfony.php and api-reference.