Symfony AI

Use Symfony AI with Eden AI to reach 500+ AI models, plus OCR, speech and image models, from one PHP platform.

Overview

Symfony AI ships an official Eden AI bridge for its Platform component. It covers both halves of the Eden AI V3 API:

The OpenAI-compatible endpoints, /v3/chat/completions and /v3/embeddings, with streaming and tool calling.
The expert models served by /v3/universal-ai: OCR, document parsing, text to speech, speech to text, image analysis and image generation.

The bridge landed in Symfony AI v0.14.0. Binary content is uploaded through Eden AI’s upload endpoint for you, and long-running transcriptions come back as a job handle rather than a blocking call.

Installation

composer require symfony/ai-eden-ai-platform

Configuration

Store your key in the environment, then declare the platform in your bundle configuration:

EDENAI_API_KEY=your_api_key_here

ai:
    platform:
        edenai:
            api_key: '%env(EDENAI_API_KEY)%'
    agent:
        default:
            model: 'openai/gpt-4o-mini'

Get your key from the Eden AI dashboard.

Quick Start

Outside a Symfony application, build the platform directly from the factory:

use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Message\Message;
use Symfony\AI\Platform\Message\MessageBag;

$platform = Factory::createPlatform($apiKey);

$messages = new MessageBag(Message::ofUser('What is the Symfony framework?'));
echo $platform->invoke('openai/gpt-4o-mini', $messages)->asText();

$vectors = $platform->invoke('openai/text-embedding-3-small', 'Some text')->asVectors();

Switching provider is a change of model string, nothing else.

Streaming

Pass stream in the options and iterate the result. Tokens arrive as they are produced:

use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Message\Message;
use Symfony\AI\Platform\Message\MessageBag;

$platform = Factory::createPlatform($apiKey);

$messages = new MessageBag(Message::ofUser('List the first 50 prime numbers.'));
$result = $platform->invoke('openai/gpt-4o-mini', $messages, [
    'stream' => true,
]);

foreach ($result->asStream() as $chunk) {
    echo $chunk;
}

Agents and Tool Calling

The platform plugs into Symfony’s Agent component, so tools work the same way they do with any other bridge. The agent decides when to call a tool, runs it, and feeds the result back to the model:

use Symfony\AI\Agent\Agent;
use Symfony\AI\Agent\Bridge\Wikipedia\Wikipedia;
use Symfony\AI\Agent\Toolbox\Toolbox;
use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Message\Message;
use Symfony\AI\Platform\Message\MessageBag;

$platform = Factory::createPlatform($apiKey);

$toolbox = new Toolbox([new Wikipedia($httpClient)]);
$agent = new Agent($platform, 'openai/gpt-4o', toolbox: $toolbox);

$messages = new MessageBag(Message::ofUser('Who is the current chancellor of Germany?'));
echo $agent->call($messages)->getContent();

Point the agent at a different model string to run the same toolbox on another provider.

OCR and Document Parsing

Pass a Document or an Image for a local file, or a DocumentUrl or ImageUrl for a remote one. A local file is uploaded through /v3/upload before the request, so both forms behave the same from your code:

use Symfony\AI\Platform\Bridge\EdenAi\DocumentParser\Result\DocumentParsingResult;
use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Message\Content\Document;
use Symfony\AI\Platform\Message\Content\DocumentUrl;

$platform = Factory::createPlatform($apiKey);

// Resume parsing from a local PDF, uploaded automatically
$result = $platform->invoke('ocr/resume_parser/openai/gpt-4o', Document::fromFile('./resume.pdf'));

$parsing = $result->asObject();
\assert($parsing instanceof DocumentParsingResult);

echo json_encode($parsing->getExtractedData(), \JSON_PRETTY_PRINT | \JSON_UNESCAPED_UNICODE);

// Invoice parsing from a URL
$result = $platform->invoke('ocr/financial_parser/affinda', new DocumentUrl('https://example.com/invoice.pdf'), [
    'language' => 'en',
    'document_type' => 'invoice',
]);

$invoice = $result->asObject();
\assert($invoice instanceof DocumentParsingResult);

Plain OCR returns raw text and bounding boxes instead of structured fields. Input parameters such as language go in the options array, or inline in the model name (ocr/ocr/google?language=en):

use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Bridge\EdenAi\Ocr\Result\OcrResult;
use Symfony\AI\Platform\Message\Content\ImageUrl;

$platform = Factory::createPlatform($apiKey);

$result = $platform->invoke('ocr/ocr/google', new ImageUrl('https://example.com/scan.jpg'), [
    'language' => 'en',
]);

$ocr = $result->asObject();
\assert($ocr instanceof OcrResult);

echo $ocr->getText();

Not every provider reads an uploaded file. ocr/financial_parser/openai returns a successful but empty result when its input is a file ID, while it parses the same document handed over as a URL. ocr/financial_parser/affinda reads both. Pass a DocumentUrl to the providers behaving that way.

Speech and Image Models

Audio, document and image content is uploaded before the request, so a local file works the same way as a URL:

use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Message\Content\Audio;
use Symfony\AI\Platform\Message\Content\ImageUrl;

$platform = Factory::createPlatform($apiKey);

// Text to speech, returned as binary data
$result = $platform->invoke('audio/tts/amazon/neural', 'Your order has shipped and arrives on Thursday.');
$result->asFile('notification.mp3');

// Speech to text from a local file, uploaded automatically
$result = $platform->invoke('audio/speech_to_text_async/deepgram', Audio::fromFile('./call.mp3'), [
    'language' => 'en',
]);
echo $result->asText();

// Object detection
$analysis = $platform->invoke('image/object_detection/google', new ImageUrl('https://example.com/photo.jpg'))->asObject();
foreach ($analysis->getItems() as $item) {
    echo $item['label'];
}

// Image generation
$result = $platform->invoke('image/generation/stabilityai', 'A product photo of a leather backpack on a white background');
$result->asFile('backpack.png');

Image analysis also covers explicit content, logo detection, face detection, AI detection and deepfake detection.

Asynchronous Jobs

Speech to text runs on /v3/universal-ai/async. When a provider answers fast enough, the transcription is in the response and the call is over. Otherwise the invocation returns a JobResult, a handle you can wait on straight away or store and pick up from a worker later:

use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Exception\JobTimeoutException;
use Symfony\AI\Platform\Job\JobRunner;
use Symfony\AI\Platform\Result\JobResult;

$result = $platform->invoke('audio/speech_to_text_async/deepgram', 'https://example.com/audio.mp3');

if ($result->getResult() instanceof JobResult) {
    $handle = $result->asJob();
    $jobClient = Factory::createJobClient($apiKey);

    try {
        $result = (new JobRunner())->wait($jobClient, $handle, maxDuration: 60);
    } catch (JobTimeoutException) {
        // The handle holds no connection, so store it and resume later
        $this->queue->dispatch(new ResumeTranscription($handle->getId()));

        return;
    }
}

echo $result->asText();

The bridge never blocks inside invoke(). How long to wait, and whether to wait at all, stays your decision. Pass a webhook_receiver option to be notified out of band instead, and resolve the handle when the webhook fires. Inside a Symfony application the job client is a service, autowirable by name:

use Symfony\AI\Platform\Job\JobClientInterface;
use Symfony\AI\Platform\Job\JobRunner;
use Symfony\Component\DependencyInjection\Attribute\Autowire;

public function __construct(
    #[Autowire(service: 'ai.platform.job_client.edenai')]
    private JobClientInterface $edenaiJobClient,
    private JobRunner $jobRunner,
) {
}

Available Models

Chat and embeddings use the provider/model format: Chat

openai/gpt-4o-mini
anthropic/claude-sonnet-5
mistral/mistral-large-latest
google/gemini-2.5-flash

Embeddings

openai/text-embedding-3-small
mistral/mistral-embed

Expert models use feature/subfeature/provider, and some take the backing model as a fourth segment:

ocr/ocr/google
ocr/financial_parser/affinda
ocr/resume_parser/openai/gpt-4o
audio/tts/amazon/neural
image/object_detection/google

Browse the full list in the Eden AI models catalog, or call List LLM Models.

Dynamic Model Catalog

The bundled ModelCatalog curates a subset of what Eden AI serves. To reach anything else, register it through the $additionalModels argument, or hand the factory a ModelApiCatalog, which discovers the current catalog from Eden AI’s public endpoints:

use Symfony\AI\Platform\Bridge\EdenAi\Factory;
use Symfony\AI\Platform\Bridge\EdenAi\ModelApiCatalog;

$platform = Factory::createPlatform($apiKey, $httpClient, new ModelApiCatalog($httpClient));

// No catalog entry needed for this one
$result = $platform->invoke('audio/tts/elevenlabs/eleven_multilingual_v2', 'Welcome!');

Expert subfeatures the bridge has no result converter for stay hidden from that catalog, so an unsupported model fails at lookup time rather than during conversion.

Runnable Examples

Symfony ships working scripts for every capability above, in examples/edenai: chat, streaming, tool calling, embeddings, OCR, invoice and resume parsing, text to speech, speech to text with and without upload, object detection, logo detection and image generation.

Next Steps

Chat Completions - Core LLM endpoint
List LLM Models - Browse available providers and models
Expert Models - OCR, speech, image and text features
Symfony AI documentation - The bridge reference on symfony.com

V3 Documentation

Quick Start

Overview

LLMs

Expert Models

General

Organization

Data Governance

Integrations

Overview

Installation

Configuration

Quick Start

Streaming

Agents and Tool Calling

OCR and Document Parsing

Speech and Image Models

Asynchronous Jobs

Available Models

Dynamic Model Catalog

Runnable Examples

Next Steps

​Overview

​Installation

​Configuration

​Quick Start

​Streaming

​Agents and Tool Calling

​OCR and Document Parsing

​Speech and Image Models

​Asynchronous Jobs

​Available Models

​Dynamic Model Catalog

​Runnable Examples

​Next Steps

Overview

Installation

Configuration

Quick Start

Streaming

Agents and Tool Calling

OCR and Document Parsing

Speech and Image Models

Asynchronous Jobs

Available Models

Dynamic Model Catalog

Runnable Examples

Next Steps