You can generate a summary of a document using a temporary list index, then add it to metadata, but only for embeddings.
for document in documents
temp_index = ListIndex.from_documents([document])
summary = temp_index.as_query_engine().query("Summarize: ")
document.metadata["summary"] = summary
document.excluded_llm_metadata_keys = ["summary"]
index = VectorStoreIndex.from_documents(documents)If you would rather do this with nodes, the same logic applies. Just change the index creation to
index = VectorStoreIndex(nodes) and similar for the list index
