Back to store

Jev-cu

Hands the "where to click next" decision in Computer Use over to Jev, reading only the interface's text candidates, never a screenshot.

Price

Free

Made by

Sac-YSac-Y@Sac-Y

Category

Agents and Routing

Type

Library

Jev-cu splits computer-use automation into two jobs: Jev picks the element, the action, whether the task is done, and how risky the step is, purely from a text description of what's on screen, while Codex Computer Use reads the actual interface and carries the action out.

It installs as a skill into Codex, with a local policy gate that blocks sensitive operations, deleting, sending, paying, authorizing, uploading, entering a code, installing, or changing a system setting, unless confirmed, and runs in dry-run mode by default.

It ships with an offline evaluation against recorded accessibility snapshots, so the element-picking accuracy can be checked without spending on live Jev calls.

What you get

  • Full JavaScript source code under the MIT license
  • The installable Codex skill, the decision loop, and an offline accuracy evaluation