Codex CLI v0.169.0 keeps side chats in tabs and saves them across restarts

Codex CLI v0.169.0 lets you keep several side chats open at once, each in its own tab. The tabs belong to the terminal workspace, which is opt-in, so you switch it on first.

Slide text

01 02 03 04 05 Side chats

openai/codex · release notes rust-v0.169.0 · #59784 · TUI test snapshot

Codex CLI v0.169.0 keeps side chats in tabs

Investig…× Fix cache× +

Tab and Shift-Tab step through your side chats, and so do the arrow keys. With a mouse, click a tab or scroll the wheel over the tab strip.

Local side chats are now saved with their conversation, so after a restart they're waiting in the sidebar. Rename one with /rename. Closing its tab keeps it in the list.

Finding your work.

Search in the agent command center used to cover only the tasks already loaded on screen. Now it scans your whole history, matching a task's name, the start of its prompt, or its working directory.

In the full-screen transcript, a new setting, Collapse turn details, folds a finished turn's commands and tool calls under its prompt. It's off by default. Answers stay in view, and search still finds the hidden output.

Providers and Cloud.

The provider picker can now add or edit a custom provider without leaving Codex. A gateway takes a name, a base URL and the variable holding its API key. Amazon Bedrock and Google Cloud get their own settings.

With no browser on hand, over SSH or on a headless Linux machine, Codex shows a link and a one-time code to enter somewhere else. It needs custom gateway onboarding switched on, and a gateway that supports device codes.

For Cloud conversations in the terminal, you can now send images, choose the model and its reasoning effort, and archive a conversation or bring it back. The Cloud connection is opt-in, on macOS and Linux.

codex exec.

In a directory you haven't trusted, codex exec now leaves the project's own configuration and MCP servers switched off. That holds even when you give the run a writable workspace sandbox.

When you do trust the repo, a new trust flag on codex exec and codex review says so. The trust is saved for later chats in that project, and the flag refuses a project you've marked untrusted.

Fixes.

When you approve a plan and choose Yes, clear context and implement, the fresh session now receives the complete plan. If that handoff fails, Codex keeps you in the session you were already in.

In the report editor, Ctrl+S now saves a diagnostic ZIP file on your own machine, and nothing goes to OpenAI. It works even when feedback uploads are disabled.

On Windows, the sandbox handles quoted cmd.exe paths, and PowerShell under an unreadable parent folder. On macOS, pasted images keep their orientation. And the terminal stops waiting on MCP servers a shared app server never reported.

If codex exec runs in your CI, I'd update today, because a checkout you never trusted can no longer start its own MCP servers. Every change is in the v0.169.0 release notes on GitHub.