Reduce AI agent token usage by 85-95%! RLM-inspired MCP proxy with field projection & regex filtering. Handle massive tool outputs efficiently. Save $$$ on LLM costs. Works with Claude, GPT-4, any MCP server.

Production-ready test-time compute optimization framework for LLM inference. Implements Best-of-N, Sequential Revision, and Beam Search strategies. Validated with models up to 7B parameters.

Implementation of Recursive Language Model paper from scratch