Caveman
Optimize tokens and reduce AI agent costs
Caveman is a local optimization layer for coding agents that reduces context before it reaches your model provider. Its proxy works with Claude Code, Codex, Gemini, Aider, Hermes, and others, compressing logs, JSON, code, diffs, and terminal output while keeping the original data recoverable. It also includes a skill that helps agents work with fewer tokens, making AI development workflows more efficient.