What it does
The examples include judging whether a coding-agent request could destroy data or expose secrets.
Integrations
Small, single-file examples that combine Pydantic AI with Jev.
Screenshot unavailable. Open the experiment ↗
The examples include judging whether a coding-agent request could destroy data or expose secrets.
Start by drafting simple questions about your app's safety rules. For example, you might ask if a user request destroys files or leaks private information. Decide what the app should do for each answer, such as running the request, stopping it, or pausing for human review.
A developer can adapt these example patterns to send prompts or commands to TypeSafe for evaluation before they run. This requires an access key to connect your software to TypeSafe. Remember that the text of each prompt is sent to this external service, and an automated answer cannot guarantee absolute safety.