A self-hosted LLM gateway that slashes token usage for AI coding tools through JSON compression, semantic caching, and complexity-based tier routing — with zero code changes.