Codex CLI v0.130 Ships Remote Control, Computer Use, and a Browser Extension

Three features in one release. Each one would've been a headline on its own.

OpenAI shipped Codex CLI v0.130 with a codex remote-control entrypoint for headless, remotely controllable agent sessions. Computer use that lets Codex see, click, and type in macOS applications. And Codex for Chrome, a browser extension that runs agents across tabs on web applications. Over 4 million people use Codex weekly.

The remote-control feature is the quiet bombshell. It turns Codex into a scriptable, headless agent you can call from CI/CD pipelines, orchestration scripts, or other agents. No TUI required. Combined with the new thread automations (wake the same thread on a schedule while preserving context), this creates persistent, programmable coding agents that behave like services rather than interactive tools.

Computer use is the answer to a problem I've hit repeatedly: GUI-only bugs. When your agent can't reproduce a rendering issue because it can only read code, you're back to manual testing. Now Codex can operate macOS apps by seeing and clicking them. It's not perfect (I haven't tested edge cases), but the capability gap it fills is real.

The Chrome extension is arguably the most user-visible addition. Running an agent that can navigate web tabs, interact with web applications, and coordinate across browser contexts opens up testing and automation workflows that were previously duct-taped together with Playwright scripts and hope.

Separately, OpenAI pushed Codex into the ChatGPT mobile app with live thread state, file diffs, terminal output, and test results on iOS and Android. Remote SSH and Hooks hit GA across all plans including Free. Your phone becomes a live agent control panel.

For builders: try codex remote-control this week. If you're running any kind of automated development workflow, headless agent sessions are how you integrate AI coding into your pipeline without the TUI overhead.