all products
Agent toolkit

Caveman Code

Talks like a caveman, ships like a senior — burns half the tokens

A terminal coding agent that uses the same models as everyone else but burns roughly half the tokens, stacking four compression layers under an autonomous goal loop. Plan first, then ship — across 20+ providers with OAuth or API keys.

npm install -g @juliusbrussee/caveman-codeGitHub480

Four layers, half the tokens

caveman-code · v0.19
$caveman-code "add a /healthz route and a test"

token budget

messages2.1k / 30k
memory1k / 40k

layers active

caveman mode
tool budgets
read dedup
memory recall

≈ half the tokens of a stock agent

What it can do

  • Four compression layers: Caveman Mode, tool budgets, read-dedup, optional RTK
  • Autonomous goal loop with autopilot, plus read-only plan mode
  • 20+ providers (Claude, ChatGPT, Copilot, Gemini, Vertex, OpenAI, Azure, Groq, DeepSeek…)
  • Architect/editor model split, session branching, shadow-git checkpoints
  • Persistent memory via cavemem, MCP servers, up to 7 worktree-isolated subagents
language
TypeScript
license
MIT