Caveman (YC F26) is now backed by Y Combinator It started as a joke: make Claude talk like a caveman so my AI bill stopped beating my grocery bill. 111,000 GitHub stars later, it's a company. And… | Julius Brussee | 216 comments
Caveman (YC F26) is now backed by Y Combinator It started as a joke: make Claude talk like a caveman so my AI bill stopped beating my grocery bill. 111,000 GitHub stars later, it's a company. And today we launched everything at once. Caveman 4.0 is out. → Say less. The original caveman voice. 63 tokens become 20. → Read less. Logs, JSON and test output shrink before the model reads them. 33.2% fewer input tokens across whole Claude Code sessions. → Auto (new). A model in your agent's picker that chooses the model and the effort for each ask. 54% cheaper on SWE-Together and SWE-Interact, same score, taking 50% less time. → Blocks (new). Agents write the same throwaway script every session. We counted one script written 394 times in a week on one machine. Blocks keeps it in your repo and hands it back. One test run: 397 tokens → 76. → Warm cache (new). Step away and your provider forgets your session, so the next message pays full price. Caveman keeps it warm while you pause. All of it i
LinkedInPerskaityta