OpenAI Codex CLI v0.149.0 Changelog

This release is all about giving you better control over your AI coding sessions. It matters most to developers who use the Codex command-line tool to manage multiple tasks at once, especially if they like using keyboard shortcuts (like Vim) or need to run AI agents in the background while they do other work.

What's new

• Agent Dashboard: There's a new interactive screen where you can search, start, open, rename, and stop AI tasks (called "agents"). You can even set up custom keyboard shortcuts for it.
• Folder Commands: You can now use /cd, /pwd, and /cwd commands right inside the text interface to see or change the folder you're working in.
• Message Queuing: A new codex queue command lets you send messages to tasks that are already running, whether they are on your computer or running remotely on another server.
• More Vim Shortcuts: If you use Vim keyboard shortcuts to edit text, you can now do more things, like replacing characters or using cw, c$, and cc to quickly change words or lines.
• Smarter codex doctor: The doctor tool (which checks if your system is healthy) can now diagnose security software blocking, network/proxy failures, desktop app issues, and problems with downloading updates.
• Better Developer Tools: For programmers using the SDK (a toolkit for building apps with Codex), you can now override exact settings and choose how much "thinking" effort the AI puts into a task (max or ultra reasoning).

Improvements

• The documentation was updated to clarify that outside volunteers should suggest changes through discussions, not direct code pull requests.
• The docs now also include warnings about DNS exfiltration risks (how hackers might steal data through network lookups) and trust limits for secure development containers.

Fixes

• Waking up tasks: Queued messages now reliably wake up idle tasks. It also fixes an issue where tasks with the same name got confused, and it makes sure copied-and-pasted commands work the way you expect.
• Restoring Permissions: When you restart or branch off an old conversation, it now correctly remembers the security permissions you had set, instead of silently resetting them to default.
• Duplicate Text: Fixed a bug where sub-agent (helper AI) activity was showing up twice on your screen, and made the notifications for them less cluttered.
• Reconnecting WebRTC: If your real-time audio/video connection drops unexpectedly, it will now automatically reconnect without losing the text or output that was being sent.
• Windows Terminal History: Fixed an issue where your command history wasn't available to scroll back through in the Windows Terminal.
• Memory Limits: Inactive background tasks used to hold onto way too much text data in memory. This is now capped so it doesn't eat up all your computer's RAM.

Breaking changes

No breaking changes in this release.