explaingit

juliusbrussee/caveman

🔥 Hot61,822PythonAudience · developerComplexity · 2/5ActiveLicenseSetup · easy

TLDR

A plugin for AI coding assistants that strips filler and makes responses ultra-terse, saving tokens and API costs while keeping technical accuracy intact.

Mindmap

mindmap
  root((caveman))
    What it does
      Terse AI responses
      Token compression
      Cost reduction
    How it works
      Plugin installation
      Multiple intensity modes
      Companion commands
    Use cases
      Faster coding feedback
      Lower API bills
      Commit message writing
      Code review summaries
    Tech stack
      Python
      Shell scripts
      JavaScript
    Supported agents
      Claude Code
      Cursor
      Windsurf
      Cline

Things people build with this

USE CASE 1

Get faster AI coding responses by removing verbose explanations and filler phrases.

USE CASE 2

Reduce API costs by compressing token usage while maintaining technical accuracy.

USE CASE 3

Write terse git commit messages and one-line code reviews using caveman-commit and caveman-review commands.

USE CASE 4

Track cumulative token savings over time with caveman-stats to measure cost reduction.

Tech stack

PythonShellJavaScript

Getting it running

Difficulty · easy Time to first run · 5min
Use freely for any purpose including commercial, as long as you keep the copyright notice.

In plain English

Caveman is a plugin and skill for AI coding assistants (primarily Claude Code, but also Codex, Gemini CLI, Cursor, Windsurf, Cline, and many others) that instructs the AI to respond in an ultra-terse, stripped-down style inspired by caveman speech. The idea, which the README claims is backed by research, is that AI models can convey the exact same technical information using far fewer words, and fewer words means fewer tokens, which means faster responses and lower API costs. The tool works by installing a plugin or skill file into your coding agent's configuration directory. Once active, the AI drops filler phrases, pleasantries, and verbose explanations, and gives you only the core technical answer. For example, instead of a paragraph explaining why a React component re-renders, you get: "New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo." The plugin offers multiple intensity levels (lite, full, ultra, and a classical Chinese mode) so you can tune how aggressively it compresses output. Beyond mode switching, the tool also provides companion commands like caveman-commit for writing terse git commit messages, caveman-review for one-line code reviews, caveman-compress for shrinking the prompts you send to the AI, and caveman-stats for tracking how many tokens you have saved over time. Installation is a single command that auto-detects which agents you have installed. You would use caveman if you are an AI-assisted developer who wants faster responses and lower costs without sacrificing technical accuracy. The stack is Python-based with shell scripts and JavaScript for the various agent integrations.

Copy-paste prompts

Prompt 1
Install caveman plugin for my coding assistant and set it to full mode so Claude gives me ultra-short technical answers without filler.
Prompt 2
Use caveman-compress to shrink the prompt I'm about to send to my AI assistant so it uses fewer tokens.
Prompt 3
Show me how to switch between caveman intensity levels (lite, full, ultra) to control how aggressively my AI compresses its responses.
Prompt 4
Run caveman-stats to see how many tokens I've saved since installing the plugin.
Prompt 5
Set up caveman-review mode so my AI gives one-line code review feedback instead of long explanations.
Open on GitHub → Explain another repo

Generated 2026-05-18 · Model: sonnet-4-6 · Verify against the repo before relying on details.