Caveman – why use many token when few do trick (github.com)

A new plugin called "Caveman" has been announced, designed for Claude Code and Codex agents, which allows these systems to generate responses using significantly fewer tokens—up to 75% less—without compromising technical accuracy. This plugin leverages the idea that simplifying language can maintain essential information while enhancing readability and speed. Key features include a variety of conversation modes, including a classical Chinese 文言文 option, and tools for generating concise commit messages and one-line code reviews. The implications for the AI/ML community are notable, as this innovation demonstrates a shift towards more efficient language model interaction. By reducing token usage, not only do users save on costs associated with API usage, but they also experience faster response times—approximately three times quicker, as less text needs to be processed. This could pave the way for future developments in language models where brevity leads to enhanced performance, challenging the notion that verbose outputs are superior. Caveman showcases how thoughtful adjustments in language processing can yield practical benefits in real-world applications.