RegressIQ

RegressIQ is a web app (with a CLI) for teams shipping LLM features who keep breaking quality without noticing. It runs scheduled, version-to-version regression tests on your prompts, tools, and model configs, then flags statistically meaningful quality drops and cost/latency regressions. Instead of generic “eval scores,” it focuses on what actually changes after a model swap, prompt edit, or tool update: refusal rate, hallucination markers, JSON validity, tool-call accuracy, and business-specific rubric checks. It supports golden sets plus live traffic sampling, so you can detect failures that only happen in the wild. Reports are built for engineers: diff views, failing examples, and a minimal “what changed” summary that can gate CI/CD. Realistically, it won’t replace deep research evals; it’s built to stop silent production regressions fast.