Open-source CLI that catches LLM quality drift before users do. Standalone binary, no Python, Node, or Docker. Every run compared against baseline automatically. Four metric pillars: factuality, format, tone, regression. NFR gates on latency/cost/coverage. Pluggable providers: Anthropic, OpenAI, Groq, Gemini, Ollama. CI-native with exit codes. MIT license, zero telemetry.
