- Introduced logic to utilize existing Document embeddings when not empty, avoiding unnecessary calls to embeddingClient.embed.
- Ensured immutability of the Document#content field by making it final.
- Implemented a default method cachedEmbed in the EmbeddingClient interface, responsible for checking existing document embeddings and adding them if absent.
- Updated all Vector Store implementations to employ cachedEmbed upon document addition.
- Ensured that embedding fields are returned with resolved documents in select Vector Store implementations for consistency.
Comment From: markpollack
I think we are going to get rid of Document having embeddings so not merging this and assigning to RC1 so it can be further discussed.
Comment From: tzolov
My opinion is that the Document#embedding field is useless (even harmful) and should be removed form the Document spec.
There is not a single use-case where this field would be used. It is is misleading to assume that it can be used a a cached vector value because there is not use case where you retrieve a Document from the VectorStore and then save the same document again.