Agent tooling
jevkit-py is an unofficial set of Python tools for Jev projects: a question linter, pytest record-and-replay, drift checks, calibration reports and benchmarks.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
Prajjwal Chittori's packages share one record format, so a saved test recording can double as a baseline for spotting drift. Replayed tests need no API key; a live client is only needed when recording new answers. It is unaffiliated with TypeSafe and, by its own description, not an SDK.
If an app asks Jev questions, these tools help its makers test it without paying for every run. Real answers are recorded once and replayed in later tests, so routine checks need no secret key. A separate report compares past answers with known right ones and suggests how confident Jev should be before an answer is trusted automatically.