Caveman

Caveman is an efficiency layer for teams that build on large language models. It sits between your application and the AI providers you use, watches the traffic, and applies cost-saving optimizations like caching, compression, and routing. The main promise is that it can cut roughly 65% of your AI costs by reducing the number of output tokens you pay for, then show you a breakdown of what was saved.

The tool is deliberately lightweight to install. There are four ways to run it: a Claude Code skill installed with a curl command, a global npm package called Caveman Code, a package called Cavemem, and a browser extension that works with ChatGPT, Claude, and Gemini. That range of entry points makes it useful inside AI-native coding agents, in backend services, and in everyday chat interfaces. The project reports about 97.9k GitHub stars and was #1 on Hacker News, which points to strong interest from developers.

Who benefits most? Engineering teams that pay significant LLM API bills, developers building AI agents, and companies that want to lower infrastructure costs without rewriting their code. Caveman is not an AI model itself; it is a cost-control layer for the AI tools you already use. The verified-savings reporting is especially useful for teams that need to justify a new tool to management.