Agent tooling
Eve's t.judge evaluates an agent's response against supplied criteria, with Jev through Vercel AI Gateway as its default decision model.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
This focused feature listing complements JevMade's existing Eve framework entry. The judge is separate from the tested agent and supports boolean, rubric-score and choice checks, including batches. Returned probabilities are estimates, not guaranteed calibrated confidence; the relevant AI SDK support is experimental.
Use an ordinary assertion for results you can check exactly. For a judgment call, supply t.judge with specific criteria and a threshold, plus the gateway credentials. A multi-turn evaluation needs the documented transcript, not only the latest response, so the judge can see what the agent was asked to do.