Hello! Ok, so I am using n8n to upsert documents into vector stores (mainly in Qdrant). I have an AWS account, and see that I can use Bedrock embeddings models. I wanted to upsert/vectorize some images so I set the model to Titan Multimodal Embeddings.
HOWEVER, the default document loader does not work for this since png is not a supported MIME Type. The only other option I have is to put in a Github repository, but since I am not a developer I really don’t know what to do there.
Does anyone know how to set up the document upsert with multimodal? Or is there a specific Git repository I can link to to make it work?
I don’t think it’s currently possible unless you can get the n8n team to enable the @aws-sdk external dependency on your instance.
Since n8n is using langchain under the hood and the langchain version only supports text embeddings, not much you can do here but wait until a future update where Bedrock multimodal embeddings are supported.
Self-hosted
You can use the “Code” node with preinstalled Bedrock SDK to get Multimodal embeddings.
Unfortunately, it’s not possible to plug “Code” node directly into the “Qdrant Vectorstore” node, so you’ll have to use the Qdrant HTTP API instead to store the embedding.
Prerequisites
You’ll need to set the environment variable NODE_FUNCTION_ALLOW_EXTERNAL=@aws-sdk