Caveman
Products
Compressionlive

Caveman Middleware

Your framework. Smaller tool results.

Keep your framework, provider client, and agent loop. Caveman Middleware compresses eligible outbound tool results through a local runtime and lets your agent retrieve the original when it needs more detail. Your application keeps its original conversation; inference stays with your provider. TypeScript and Python packages are available in alpha.

Capability ledger

06
  1. 01Native adapters for existing TypeScript and Python agent frameworks
  2. 02Compresses supported tool-result content before the model call
  3. 03Registers original-recovery tools for supported agent loops
  4. 04Preserves your application's original conversation history
  5. 05Local compression runtime; inference stays with your model provider
  6. 06Framework-specific quickstarts, supported version ranges, and diagnostics

Product ledger

02
Language
TypeScript · Python
License
See package licenses

Product index

10
01Caveman SkillThe MIT skill: output compression for Claude Code & 30+ agents.Compression / live02Caveman ProxyRecoverable local context compression for agents you already use.Compression / live03Caveman BrowseToken-efficient browser automation: compressed a11y snapshots, byte-exact recovery.Agent toolkit / live04CaveGemmaCaveman compression baked into Gemma's weights.Compression / live05Caveman PlatformLower cost per successful agent task, with quality and reliability as constraints.Cloud / in development06RouterEval-gated model routing: the most efficient model in your pool that passes your evals.Cloud / in development07PebbleA coding agent. 3/3 tasks passed with 31.2% less estimated API spend in a three-task pilot. In private development.Agent toolkit / in development08Caveman CodeTerminal coding agent. Half the tokens.Agent toolkit / live09CavememPersistent memory your agents recall over MCP.Agent toolkit / live10CavekitCompressed, spec-driven development.Agent toolkit / live