How are different embeddings compatible?

In RAG we use:
Embedding model at the input, it can be stored somewhere but how are these embeddings interpreted by LLM later on? Does it not have a different embeddings (even if they are the same size those vectors could have different meanings).

Could someone please clarify whether LLM is receiving embeddings from the embeddings model used to create entries to VectorDB or does it receive converted text from embeddings?

Thanks,
Jed