Jev in production › Agent guardrails and approvals
Classifies shell commands with Jev, or local Kev/Laya, to auto-run, confirm, or block tool calls before execution in the Pi coding agent.
An auto mode plugin for the Pi coding agent that uses a classifier model, Jev (hosted) or Kev/Laya (running locally), to classify shell commands before they run.
It decides per tool call whether the call runs, needs a confirm prompt, or is blocked. Built-in rules decide most calls. The model checks the shell commands the rules don't know, and you get a prompt for the risky ones.
I built it because Pi runs every tool call without asking for approval, and I wanted something like Claude Code's auto mode. With Kev or Laya on CPU it works fully offline and doesn't use the GPU. With Jev the commands are sent to the provider, so that one is opt-in.
For the project's own README, linking back here: