AI代理在你的导航栏上浪费了90%的令牌——我建立了一个协议来解决这个问题

Users have found that a significant portion of their spending on AI API calls does not yield useful information, as it often involves processing irrelevant elements like navigation bars, cookie banners, and inline scripts. For instance, a typical e-commerce product page may involve 47,000 tokens, with only about 300 tokens containing useful product information. This indicates that over 90% to 99% of the processed tokens are structural or navigational noise. To address this issue, the MAKO (Markdown Agent Knowledge Optimization) protocol has been proposed, which employs standard HTTP content negotiation to deliver AI-optimized content, aiming to provide AI agents with the necessary information directly, thereby enhancing efficiency and reducing costs.