Skip to content
Stork.AIstork.ai
AI Tool

caveman Review

Caveman is an AI agent skill that cuts approximately 75% of output tokens by making agents communicate in a terse, compressed style, while maintaining technical accuracy.

shipped Jun 24, 2026freemium
Monthly visits3/mo
caveman - AI tool for caveman. Professional illustration showing core functionality and features.

Why it matters

1Reduces AI token consumption by 65-75%, leading to substantial cost savings.
2Achieves approximately 3x faster responses for AI agent interactions.
3Accumulated over 76,000 GitHub stars since its release in April 2026.
4Supports output compression for Claude Code, Gemma, ChatGPT, Gemini, and 30+ other AI agents.

Specs

GitHub

View Repository →

overview

What is caveman?

caveman is a token optimization tool developed by Julius Brussee that enables AI developers and organizations using AI agents to significantly reduce token usage and associated costs. It achieves this by instructing AI models to respond in a terse, compressed style while preserving technical accuracy. Caveman's core function is to optimize Large Language Model (LLM) interactions by removing filler words, politeness, and unnecessary explanations, focusing solely on core technical meaning. This 'semantic compression' aims to reduce token consumption, leading to lower API costs and faster response times. The tool operates as a skill or plugin for various AI models and platforms, including Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, and Copilot. It can be activated in a session using commands like /caveman and offers three intensity levels: Lite, Full (default), and Ultra, alongside a 文言文 (Classical Chinese) mode.

features

Key Features of caveman

Caveman provides a suite of features designed to optimize AI agent interactions by reducing token consumption and enhancing efficiency across various development workflows. These capabilities extend beyond simple output compression to include persistent memory solutions and a comprehensive LLM gateway.

  • AI agent skill for token reduction, achieving 65-75% compression.
  • Output compression for Claude Code and over 30 other AI agents.
  • CaveGemma, a specialized compression method baked directly into Gemma's weights.
  • Caveman Code: A terminal coding agent designed for approximately 50% token reduction.
  • Cavemem: Persistent memory for AI agents, utilizing local SQLite with FTS5 and a vector index.
  • Cavekit: A framework for compressed, spec-driven development.
  • Caveman Proxy: A byte-safe LLM gateway for transparently managing any LLM traffic.
  • Browser extension for ChatGPT, Claude, and Gemini, enabling one-click 'Caveman Mode'.
  • Cave Architect: A telemetry system for ranked plan and cost analysis of LLM interactions.
  • Eval-Gated Rollout: A system for verified savings through replay, shadow, canary deployments, and auto-rollback.

use cases

Who Should Use caveman?

Caveman is primarily targeted at AI developers, organizations, and individual users who seek to optimize their interactions with Large Language Models by reducing operational costs and improving response efficiency. Its design addresses specific pain points related to token consumption and verbose AI outputs.

  • AI developers: To significantly reduce token consumption and associated costs for AI models, particularly in scalable agent systems.
  • Organizations using AI agents: For optimizing AI output by stripping away linguistic filler and pleasantries, leading to cheaper and faster automated pipelines.
  • Users of AI agents (e.g., for coding and debugging): To maintain full technical accuracy while reducing verbosity, receiving faster and cleaner fixes (e.g., 'Auth bug. Expiry check wrong. Use < not <= .').
  • Developers requiring memory optimization: For compressing project memory files (e.g., CLAUDE.md) to reduce recurring input token costs across sessions by an average of 46%.
  • Teams needing concise communication: For enforcing brevity in conventional commit messages (under 50 characters) and one-line pull request comments (e.g., 'L42: 🔴 bug: user null. Add guard.').

pricing

caveman Pricing & Plans

Caveman operates on a freemium model, providing core functionalities without charge. This allows users to access key token reduction and agent enhancement tools for free, with potential for future premium offerings or managed services like Caveman Cloud.

  • Freemium: Free (includes Claude Code skill, Caveman Code, Cavemem, and the Browser extension).

Similar Tools

caveman vs Competitors

Caveman positions itself as a 'compression primitive' and a 'token-efficient stack' for agent-native development, primarily focusing on reducing output token usage without sacrificing technical accuracy. Its approach differs from competitors that often focus on input prompt optimization or algorithmic compression.

1
LLMLinguaOn Stork

LLMLingua uses a smaller language model to algorithmically compress prompts by identifying and removing non-essential tokens, achieving significant compression ratios.

Unlike 'caveman' which applies a stylistic transformation to Claude's output to reduce tokens, LLMLingua focuses on compressing the input prompt before it reaches the LLM. It is a developer-focused library for programmatic compression.

2
RTK (Rust Token Killer)On Stork

RTK is a CLI proxy that intercepts and filters verbose terminal output before it reaches AI coding agents like Claude Code, significantly reducing input token consumption.

While 'caveman' optimizes Claude's generated responses by making them terse, RTK optimizes the data fed *into* Claude from terminal commands. Both target token reduction within a coding environment.

3
Promptly↗

Promptly is a Chrome extension offering a one-click prompt optimizer that rewrites prompts for improved structure, clarity, and context, alongside prompt management features.

Promptly provides a broader suite of prompt engineering tools, including optimization that implicitly reduces tokens through better phrasing, and works across various LLMs including Claude. 'caveman' is a specific Claude Code skill focused solely on output token reduction via a unique stylistic change.

4
Aiven Prompt Optimizer↗

Aiven Prompt Optimizer is a free online tool designed to rewrite LLM prompts to enhance clarity, structure, and token efficiency.

This tool directly competes by optimizing the input prompt for cost and clarity, aiming for conciseness. 'caveman' achieves token reduction by altering the conversational style of Claude's output, rather than modifying the initial prompt structure.

Visit caveman↗
Connect
⌘
GitHubgithub.com/JuliusBrussee/caveman