{% extends "base.html" %}{% block title %}Settings{% endblock %}{% block body %}
Settings
General
Jev second opinion
Jev (TypeSafe) reads each operation that is about to ask you and answers what it does and whether the agent's stated reason fits. Commands, paths, tool arguments and the reason are sent to TypeSafe; resource names are not, and values that look like credentials stop the call. Nothing is sent for operations already allowed, denied or refused.
{% if not jev_configured %}
No key. Set KW_TYPESAFE_API_KEY in the broker's environment to enable this.
{% endif %}
{% if jev_tally.verdicts %}
Shadow so far: {{ jev_tally.verdicts }} verdicts ({{ jev_tally.no_verdict }} without an answer), {{ jev_tally.safe }} judged safe. Would have auto-approved {{ jev_tally.would_auto }}: you approved {{ jev_tally.would_auto_approved }} and denied {{ jev_tally.would_auto_denied }} of those. Judged unsafe but approved by you: {{ jev_tally.unsafe_approved }}.