Agent tooling
Eve's t.judge checks an agent's response against your criteria, using Jev through Vercel AI Gateway as the default decision model.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
This feature has its own listing alongside JevMade's broader Eve framework page. The judge is separate from the agent under test and can run boolean, rubric-score or choice checks, including batches. Its probabilities are estimates, not guaranteed calibrated confidence. The relevant AI SDK support remains experimental.
Keep an ordinary assertion for a result you can check exactly. When judgment is needed, give t.judge specific criteria, a threshold and gateway credentials. For a conversation with several turns, supply the documented transcript rather than just the last response, so the judge can see the task the agent received.