README

AI Vector search integration for EXT:seal based on symfony/ai.

Enables semantic (vector-based) search in TYPO3 by bridging the SEAL search abstraction with the Symfony AI platform and store components. Documents are automatically vectorized via configurable embedding models and stored in a vector database of your choice.

This extension was funded by the TYPO3 Association: community ideas I / community idea II & first blogpost / second blogpost

Requirements

• PHP 8.3+
• TYPO3 13.4 or 14.2
• EXT:index & EXT:seal installed and configured

Installation

composer require lochmueller/seal-ai

Additionally, install the symfony/ai packages for your chosen platform and store (see Supported Platforms and Supported Stores).

Configuration

1. Set the SEAL search adapter to ai:// in your site configuration.
2. Configure the two DSN fields added to the TYPO3 site configuration module:
   • AI Platform DSN (sealAiPlatformDsn) — connection to the embedding/AI provider
   • AI Store DSN (sealAiStoreDsn) — connection to the vector store

The platform DSN must include a model query parameter specifying the embedding model:

openai://YOUR_API_KEY@default?model=text-embedding-3-small

DSN Format

scheme://user@host:port?param1=value1&param2=value2

• scheme — provider identifier (e.g. openai, qdrant)
• user — API key (where applicable)
• host / port — server address
• Query parameters — provider-specific options

Supported Platforms

Each platform requires its own composer package. Install only what you need.

Supported Stores

For mariadb and postgres stores, the extension reuses the existing TYPO3 database connection automatically.

Extending via Events

Both factories support the event:// DSN scheme, allowing you to provide custom platform or store implementations via PSR-14 event listeners.

• Lochmueller\SealAi\Event\PlatformFactoryEvent — dispatched for event:// platform DSN
• Lochmueller\SealAi\Event\StoreFactoryEvent — dispatched for event:// store DSN

Example event listener:

use Lochmueller\SealAi\Event\PlatformFactoryEvent;

final class CustomPlatformListener
{
    public function __invoke(PlatformFactoryEvent $event): void
    {
        $dsn = $event->getDsn();
        // Build your custom platform...
        $event->setPlatform($myCustomPlatform);
    }
}

Architecture

The extension implements the CmsIg\Seal\Adapter\AdapterInterface with these components:

• AiAdapterFactory — registered as SEAL adapter factory for the ai:// scheme, initializes the bridge from the current site configuration
• AiBridge — central service that holds the initialized platform, store, and vectorizer
• AiSchemaManager — manages index creation/deletion in the vector store
• AiIndexer — vectorizes documents and stores them; handles save, delete, and bulk operations
• AiSearcher — performs semantic search by vectorizing the query and querying the store

Development

# Install dependencies
composer install

# Fix code style (PHP-CS-Fixer, PER-CS 3.0)
composer code-fix

# Static analysis (PHPStan level 8)
composer code-check

# Run unit tests
composer code-test

# Tests with coverage (requires Xdebug)
composer code-test-coverage

Comparing available symfony/ai packages

diff --color \
  <(grep -E 'symfony\/ai-[a-z0-9-]+-(store|platform)' composer.json) \
  <(curl -s https://raw.githubusercontent.com/symfony/ai-bundle/refs/heads/main/composer.json \
  | sed -nE '/"symfony\/ai-[a-z0-9-]+-(store|platform)":/s/^    //p')

Updating available symfony/ai packages

jq --argjson updates "$(curl -sL https://raw.githubusercontent.com/symfony/ai-bundle/main/composer.json | jq '."require-dev" | with_entries(select(.key | test("^symfony/ai-.*-(store|platform)$")))')" '."require-dev" |= ((. // {}) + $updates | to_entries | sort_by(.key) | from_entries)' composer.json | sponge composer.json