Compress tool outputs, logs, files, RAG chunks before LLM for 60-95% less tokensgithub.com/chopratejas4 pointsgmays3 days ago