Entroly is a local-first proxy designed to cut AI coding bills without sacrificing answer quality. It works as a drop-in replacement for AI coding tools, compressing context and optimizing token usage by 70-95%. The tool includes a hallucination guard that achieved 0.844 AUROC on HaluEval-QA benchmarks at no additional cost. Built with Rust and WASM for performance, it supports 34+ integrations including Claude, OpenAI, Gemini, Cursor, Claude Code, Codex, and Aider.