Caveman
Products
Compressionlive

Caveman

Why use many token when few token do trick

A skill and plugin for Claude Code, Codex, Gemini, Cursor, and 30+ agents. It removes filler from model replies while keeping code, commands, and errors byte-for-byte exact. The 65% benchmark covers output only; input and reasoning tokens are unchanged.

Product demo / Caveman

illustrative local workbench
Compression workbench local demo
intensity
156 est. tok
Output · caveman49 est. tok

Auth: validate email pre-insert; reject disposable domains; keep `oauth_callback`; log `organization_id` + `request_id`; omit query/secrets; test valid/malformed/duplicate; return `invalid_email`.

illustrative rewrite · not production engine output

Token map · first 48 kept removed code
01voiceactive
02structuralavailable
03CCRunused
04recoverynot needed

Capability ledger

06
  1. 0165% average output-token reduction across 10 prompts (22–87% range)
  2. 02Output only: input and reasoning tokens are unchanged, and the skill adds prompt overhead
  3. 03Four intensity levels: lite, full, ultra, and wenyan (classical Chinese)
  4. 04Works across 30+ coding agents from one skill
  5. 05Byte-exact recovery for engine transformations through the CCR store
  6. 06Structural compressors for JSON, logs, ASTs and tool schemas

Product ledger

02
Language
Claude Code skill
License
MIT

Product index

05
01CaveGemmaCaveman compression baked into Gemma's weights.Compression / live02Caveman CodeTerminal coding agent. Half the tokens.Agent toolkit / live03CavememPersistent memory your agents recall over MCP.Agent toolkit / live04CavekitCompressed, spec-driven development.Agent toolkit / live05Caveman ProxyThe byte-safe LLM gateway.Cloud / in development