Different embedding model per index

For “Advanced queries with agents”: How to handle the case where two indices were created with different embedding models in a separate process? How would I know which embedding model was used for creating the vector store during retrieval?


To deal with indices that use different embedding models, you should store information about each index, like the name of the model. During search, use user input or question analysis to find the right index. To find the right embedding model, you can use the saved metadata. Then, vectorize the question and do a similarity search against the chosen vector store. In this case, tools like LangChain and Faiss can be very helpful. Remember that correct retrieval rests on matching the embedding model of the question with the model of the index.