Both use treesitter and create knowledge graphs for llm use. It results in way less tokens spent as well.