I chunk the documents and use Elasticsearch to store the vectors. On a laptop with 8GB of GPU I can have a pretty large context window and not hallucinate.