It provides a typed prompt-and-judge SDK, CLI and evidence viewer for agent evaluation, consolidating scattered scoring scripts into a reproducible pipeline. 57 stars in 1.6 days reflect evaluation moving from demos to engineering. Open source plus cloud result hosting and CI integration subscriptions could monetize it; the risk is Braintrust, LangSmith and Inspect already occupying the space, and platforms shipping their own evals.
Typed prompt-and-judge SDK, CLI, and evidence viewer for agent evaluations
★57