Caveman
Products
Compressionin development

Caveman Engine

One command wraps your agent. Prompts never leave your machine

The closed-source compression engine underneath the skill. `caveman wrap claude` proxies your agent on your machine, byte-safe; with a free account it compresses context before tokens cost you anything, and every original stays recoverable byte-for-byte. For enterprises, the same engine runs in your cloud or on-prem with zero data retention. In private development.

Capability ledger

06
  1. 01One-command wrap for Claude Code, Codex, Cursor and other coding agents
  2. 02Byte-safe by default: record mode is pass-through; on any parse problem the body goes through unchanged
  3. 03Compresses tool output, logs, JSON, diffs and stale context on-device; CCR keeps originals byte-recoverable
  4. 04Without an account it observes and reports what compression would have cut, measured locally (inferred)
  5. 05Free plan is 1 real seat with an account, no card; telemetry is token counts, never prompt bytes
  6. 06Enterprise: the engine in your cloud or datacenter — zero data retention, enforced at write time

Product ledger

03
Language
Go
License
Commercial
Pricing
Free 1 seat with an account · Indie $19/mo · Team $299/mo, 10 seats · Enterprise by contract (on-prem, zero data retention).

Product index

06
01Caveman SkillThe MIT skill: output compression for Claude Code & 30+ agents.Compression / live02CaveGemmaCaveman compression baked into Gemma's weights.Compression / live03Caveman GatewayThe byte-safe LLM gateway.Cloud / in development04Caveman CodeTerminal coding agent. Half the tokens.Agent toolkit / live05CavememPersistent memory your agents recall over MCP.Agent toolkit / live06CavekitCompressed, spec-driven development.Agent toolkit / live