Braintrust

Braintrust focuses on measuring whether your AI app is actually good. You define test cases and scoring functions, run experiments when you change a prompt or model, and see exactly what got better or worse. The same tool traces production requests and lets you turn interesting ones into new test cases, closing the loop between real use and testing. A playground and prompt versioning round out the workflow. Braintrust has a free tier and paid plans, with self-hosting for enterprises. It is popular with teams that want evaluation to be a daily habit.