Will require a few things, mainly the onus will be on the user to have a reasonable encoding system e.g batching etc. However there is value in their custom code and may have actual applications if we move it out of just the ColBERT paradigm
Simple starting point for Chain Memories, Simply take your corpus encode it and fire the whole thing into a faiss index, only a lightweight wrapper for creation and search is needed, crucially should implement a search logic