prism-orm / embedding-openai

v1.0.0-RC1 2026-10-29 09:32 UTC

Requires

• php: >=8.4
• prism-orm/orm: v1.0.0-RC1
• symfony/http-client-contracts: ^3.5

Requires (Dev)

• symfony/http-client: ^7.4 || ^8.0

Suggests

• symfony/http-client: An implementation of HttpClientInterface

MIT

Published on Packagist.org on 2026-10-29 13:47 UTC

README

The embedding provider of Prism for the POST /v1/embeddings API in the OpenAI format. The same format serves OpenAI, Mistral, Ollama (/v1), vLLM, LocalAI and Text Embeddings Inference.

composer require prism-orm/embedding-openai symfony/http-client

use Prism\Embedding\OpenAi\OpenAiCompatibleProvider;
use Symfony\Component\HttpClient\HttpClient;

$provider = new OpenAiCompatibleProvider(
    HttpClient::create(),
    baseUrl: 'http://ollama:11434/v1',
    model: 'nomic-embed-text:v1.5',
    dimensions: 768,
    apiKey: null,     // 'sk-…' for OpenAI or Mistral
    batchSize: 64,    // texts per request
);

The provider sends the texts in batches, one request per batch, and checks the number and the dimension of the vectors it receives. An HTTP error, a network error or an unusable response raises Prism\Orm\Exception\EmbeddingException: the rows stay stale, and a ComputeEmbedding message is retried.

In a Symfony application, the bundle configures it:

prism:
    embeddings:
        providers:
            default:
                openai: { base_url: 'http://ollama:11434/v1', model: 'nomic-embed-text:v1.5', dimensions: 768 }

This package is a read-only split of the Prism monorepo: open issues and pull requests there.