Caveman is a developer-centric tool designed to combat "token bloat" in AI-driven coding sessions. Inspired by the viral "why use many word" meme, it acts as a system-level prompt or plugin that strips away articles, filler words, and pleasantries from AI responses. By transforming verbose explanations into telegraphic, high-density fragments, it significantly increases response speed and reduces API costs. The project includes a dedicated "Caveman Compress" utility that even rewrites memory files (like CLAUDE.md) to save input tokens. Built with support for a wide range of agents—including Claude Code, Cursor, Windsurf, and Gemini Caveman maintains full technical integrity while proving that in high-frequency development cycles, less word is often more correct.
Key Features
-
Multi-Level Token Compression: Choose between "Lite," "Full," and "Ultra" modes to control the intensity of verbal compression and token savings.
-
Broad Agent Support: Native plugins and skills available for Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Copilot, and Cline.
-
Caveman Compress Utility: Automatically rewrite documentation and memory files into compressed prose to reduce session-start input tokens.
-
Integrated "Caveman-Commit": Generate ultra-terse, conventional commit messages directly from your AI agent.
-
One-Line Code Reviews: Perform rapid-fire PR reviews that pinpoint bugs and suggest fixes in single-line, fluff-free comments.
-
文言文 (Wenyan) Mode: Utilize Classical Chinese literary compression for even more extreme token efficiency in written responses.
Use Cases
-
High-Frequency Debugging: Speed up iterative bug-fixing sessions by eliminating the time the AI spends generating conversational filler.
-
API Cost Optimization: Dramatically lower the cost of using high-tier LLMs by cutting output token generation by an average of 65%.
-
Local Development Workflows: Improve the snappiness of terminal-based AI agents on slower connections by reducing the volume of data transmitted.
-
Technical Documentation Review: Use the "Compress" tool to shrink large instruction files into high-density reference guides that AI can ingest more cheaply.
-
No-Nonsense Code Audits: Get straight to the technical vulnerabilities and logic errors in a pull request without "throat-clearing" introductory text.
