Creating a vector store connector 

Responsibilities 

• expose a stable connector identifier;
• write vector documents;
• search vectors;
• delete all chunks belonging to a source hash;
• create or validate collections if supported;
• preserve metadata payloads;
• map backend responses to core DTOs.

Identifier 

The identifier is stored in the vector store connection record.

public function getIdentifier(): string
{
    return 'vendor.vector_store';
}

Required capabilities 

A vector store connector should support at least:

• collection-aware writes;
• similarity search with configurable max_retrieval_results and score threshold;
• payload metadata storage;
• deletion by source hash;
• stable document IDs.

Deletion by source hash 

Deletion by source hash is required for re-indexing and cleanup. The connector must delete all vector chunks that belong to the same source in the selected collection and vector store.

use Madj2k\AiCore\Connection\Configuration\VectorStoreConnectionConfigurationInterface;
use Madj2k\AiCore\Connection\VectorStore\DTO\VectorCollection;
use Madj2k\AiCore\Connection\VectorStore\DTO\VectorDeleteResult;

public function deleteBySourceHash(
    VectorStoreConnectionConfigurationInterface $connection,
    VectorCollection $collection,
    string $sourceHash
): VectorDeleteResult {
    // Delete all points with payload.source_hash = $sourceHash.
}

Service registration 

services:
  Vendor\Extension\Connection\VectorStore\MyVectorStoreConnector:
    tags:
      - name: 'aiassistant.connection.vector_store_connector'

Configuration usage 

After registration, create a vector store connection record with connector_identifier set to the connector identifier.

Example:

connector_identifier = vendor.vector_store