caveman

Tech stack & purpose

Caveman is an open-source project that reduces token usage for AI coding agents by up to 65% through specialized compression and routing techniques. The project provides a gateway system that routes multiple AI coding agents—including Claude, Codex, Gemini, Aider, OpenCode, Hermes, and OpenClaw—through unified interfaces while managing token efficiency via agent profiles, compression tools, and browser interaction capabilities. Built in Go and Node.js/TypeScript, caveman includes components for managing agent routing through environment variables and config files, compressing command output and accessibility trees, recovering original payloads through CCR (Caveman Compression and Recovery), and evaluating cache performance across different LLM providers including Anthropic, OpenAI, Bedrock, and Gemini. The project emphasizes fail-closed validation, exact token accounting, and reproducible compression without external dependencies in its CLI layer.

Languages

Contributors (top 30 of 38)

• Julius Brussee195 commits
• 
  github-actions[bot]11 commits · 0 followers
• sebastianbreguel9 commits
• AmirF1948 commits
• vraj5 commits
• Matthias Humt4 commits
• Leo3 commits
• Arkadiusz Gładki3 commits
• Akshay Patil3 commits
• Shiva Vinodkumar3 commits
• tuanaiseo3 commits
• Jimmy CrakCrn2 commits
• lbzepoqo2 commits
• cbarrado2 commits
• Ousama Ben Younes2 commits
• Test User1 commits
• David Young1 commits
• Andrew Qu1 commits
• Nishant Srivastava1 commits
• David1 commits
• Paul Jobson1 commits
• Jayesh Patel1 commits
• Matt Van Horn1 commits
• Dennis Neufeld1 commits
• Dmitrii Malakhov1 commits
• Joao Vasques1 commits
• Leonardo Colman Lopes1 commits
• fuleinist1 commits
• Fezan Muhammad Ali1 commits
• Ijtihed Kilani1 commits