ParetoOps runs in your CI to assess AI changes using the eval results you already produce. It checks every pull request against accuracy and cost limits, and with Pro compares against your last good run to flag regressions with confidence intervals. ParetoOps ranks models and prompts by cost per successful task, recommends the cheapest option that meets your bar, and works offline without calling model APIs. It supports Promptfoo, LangSmith, Braintrust, DeepEval, CSV, and JSON.
