Change embedding model for "Generate chat response" action in Retool AI

[cosmick]

Hello

I've been trying to figure out why my Retool AI conversations never include any vector documents despite selecting the checkbox to use Retool vectors:

After spending some time debugging, I noticed in the state that the embeddingModel used for the Retool AI query is set to text-embedding-ada-002, but my chosen vector is using text-embedding-3-large.

I can't confirm, but my assumption is that this is the reason why I can never get the vector context to populate in my queries no matter how specific I make my questions in relation to the vector documents.

Is it possible to add support for changing the value of embeddingModel or have Retool automatically choose the correct model based on the vector chosen?

[Mike_M]

Hi @cosmick, I brought this up with our engineers and yes a different embedding model of the query vs the document would lead to bad retrieval results.

I created a ticket to add a field in the AI query editor to change the embedding model of the query, so that we can match these up in the future

[cosmick]

Thanks, @Mike_M!

It would be great to get that added. Right now, those chat conversations aren't half as helpful as they could be with the right vector context included