Yii2 Ollama Component

Yii2 component for Ollama API with optional vector database support (e.g., Qdrant) for Retrieval-Augmented Generation (RAG).

Features

• Connect to Ollama API (llama2, mistral, gemma)
• Optional vector DB integration for context injection
• Supports Yii2 HTTP Client
• Easy configuration via Yii2 components
• Multilingual exception messages (Yii::t())
• Events support: beforeGenerate, afterGenerate, generateError
• Request and User automatically included in events for easy logging and auditing

---

Installation

Install via Composer:

composer require strtob/yii2-ollama

Ensure you have yiisoft/yii2-httpclient and a Qdrant PHP client installed if you want vector DB support.

Migration:

'controllerMap' => [
    'migrate' => [
        'class' => 'yii\console\controllers\MigrateController',
        'migrationNamespaces' => [
            'strtob\yii2Ollama\migrations',
        ],
    ],
],


yii migrate

Configuration

Example config/web.php using a Qdrant adapter:

use strtob\yii2Ollama\QdrantAdapter;

$qdrantAdapter = new QdrantAdapter($qdrantClient, 'my_collection');

'components' => [
    'ollama' => [
        'class' => 'strtob\yii2Ollama\OllamaComponent',
        'apiUrl' => 'http://localhost:11434/v1/generate',
        'apiKey' => 'MY_SECRET_TOKEN',
        'model' => \strtob\yii2Ollama\OllamaComponent::MODEL_MISTRAL,
        'temperature' => 0.7,
        'maxTokens' => 512,
        'topP' => 0.9,
        'vectorDb' => $qdrantAdapter, // must implement VectorDbInterface
        'vectorDbTopK' => 5,
    ],
];

Usage in Controller

try {
    $prompt = "Explain RAG with vector DB.";

    $text = \Yii::$app->ollama->generateText($prompt);

    echo $text;

    //or

    print_r(generateTextWithTokens(string $prompt, $options = []))

} catch (\yii\base\InvalidConfigException $e) {
    echo Yii::t('yii2-ollama', 'Configuration error: {message}', ['message' => $e->getMessage()]);
} catch (\strtob\yii2Ollama\OllamaApiException $e) {
    echo Yii::t('yii2-ollama', 'API request failed: {message}', ['message' => $e->getMessage()]);
}

---

Stream Modus

See example:

 public function actionIndex()
    {
        $response = Yii::$app->response;
        $response->format = \yii\web\Response::FORMAT_RAW;
        Yii::$app->response->isSent = true;

        // Output-Puffer leeren
        while (ob_get_level())
            ob_end_clean();
        ob_implicit_flush(true);

        Yii::$app->ollama->stream = true;

        try {
            Yii::$app->ollama->generate("Whats up?", [], function ($chunk) {
                echo $chunk;
                flush();
            });
        } catch (\Throwable $e) {
            echo "\n[Error]: " . $e->getMessage();
        }

    }

Yii2 Ollama Component – Events

OllamaComponent supports three main events during generation:

Event	When Triggered	Data Included
beforeGenerate	Before sending a request to the Ollama API	prompt, options, request, user
afterGenerate	After receiving a successful response	prompt, options, request, user, response
generateError	When an exception occurs during generation	prompt, options, request, user, exception

Event Data Details

• Connect to Ollama API (llama2, mistral, gemma)
• Optional vector DB integration for context injection
• Supports Yii2 HTTP Client
• Embedding generation (embedText)
• Easy configuration via Yii2 components
• Multilingual exception messages (Yii::t())
• Events support: beforeGenerate, afterGenerate, generateError
• Request and User automatically included in events for easy logging and auditing

---

Example Usage

// Log after generation
\Yii::$app->ollama->on(\strtob\yii2Ollama\OllamaComponent::EVENT_AFTER_GENERATE, function($event) {
    Yii::info("Prompt generated: {$event->data['prompt']}", 'ollama');
    Yii::info("User: " . ($event->data['user']->username ?? 'guest'), 'ollama');
});

// Handle errors
\Yii::$app->ollama->on(\strtob\yii2Ollama\OllamaComponent::EVENT_GENERATE_ERROR, function($event) {
    Yii::error("Generation failed for prompt: {$event->data['prompt']}", 'ollama');
    Yii::error("Exception: " . $event->data['exception']->getMessage(), 'ollama');
});

---

Embedding Generation Usage

try {
    $text = "Llamas are members of the camelid family.";
    
    $embeddingResult = \Yii::$app->ollama->embedText($text);
    
    echo "Embedding vector:\n";
    print_r($embeddingResult['embedding']); // Nur die Embeddings anzeigen

} catch (\yii\base\InvalidConfigException $e) {
    echo "Configuration error: " . $e->getMessage();
} catch (\strtob\yii2Ollama\OllamaApiException $e) {
    echo "Embedding request failed: " . $e->getMessage();
}

---

Document Model (ActiveRecord with Vector DB)

You can use an ActiveRecord model to handle documents and automatically generate embeddings for them in your vector database. Supports PDF, TXT, DOCX uploads.

use app\models\DocumentModel;
use yii\web\UploadedFile;

// 1) Create a new document
$doc = new DocumentModel();
$doc->title = 'Sample PDF';
$doc->user_id = 1;
$doc->uploadedFile = UploadedFile::getInstance($model, 'uploadedFile');
$doc->save(); // extracts text and stores embeddings automatically

// 2) Update document content and embeddings
$doc = DocumentModel::findOne($id);
$doc->title = 'Updated Title';
$doc->uploadedFile = UploadedFile::getInstance($model, 'uploadedFile'); // optional
$doc->save(); // embeddings updated automatically

// 3) Delete document and corresponding vectors
$doc = DocumentModel::findOne($id);
$doc->delete(); // deletes vectors in vector DB automatically

How It Works

• beforeSave() – Converts PDFs to text or reads uploaded TXT/DOCX.
• afterSave() – Generates embeddings using VectorizerHelper and stores them in the vector database.
• afterDelete() – Removes corresponding vectors from the vector database.

This makes your document storage fully RAG-ready, automatically connecting your database records with vector embeddings.

---

Vector DB Support

Implement the VectorDbInterface to use any vector database. Example Qdrant adapter:

use strtob\yii2Ollama\VectorDbInterface;
use strtob\yii2Ollama\QdrantAdapter;

$qdrantAdapter = new QdrantAdapter($qdrantClient, 'my_collection');

OllamaComponent will automatically prepend top-K context from the vector DB to the prompt.

---

Supported LLM Models so far

• llama2
• mistral
• gemma

---

License

MIT License – see LICENSE