Hello.

In my previous POC, I did everything manually. And it worked well! Thanks for that.

With the bundle, I discover the Indexer stuff.

It looks good.

But It seems I cannot use it. The indexer need a loader to work. I can override it with a specific source. But my use case does not fit.

I consume messages from Rabbitmq through messenger. So the fetching documents from loader part is not useful for me. But the next bits are really cool (filters, transformers, vectoriser, and store).

So I would expect a service where I'm able to do:

#[AsMessageHandler()]
final readonly class VectorizeCrawlUrlHandler
{
    public function __construct(
        // #[Target('crawl_url')]
        private IndexerInterface $indexer,
    ) {
    }

    public function __invoke(VectorizeCrawlUrl $crawlUrl): void
    {
        $this->indexer->index(new TextDocument(
            id: new Uuid($crawlUrl->id),
            content: $crawlUrl->body,
            metadata: new Metadata([
                // My stuff here
            ]),
        ));
    }
}

What do you think about this use case?

Could we consider splitting the indexer? Or we could change a bit the API to accept a document, or a collection of document. And in such situation, we skip the loader?