Codex CLI 0.167.0: an agent command center that flags unread results, no daemon required

OpenAI shipped Codex CLI 0.167.0 on November 6, 2026, and the release is about managing agent work, not a new model. The headline feature is an agent command center that flags results you haven't read yet — and you can open it without a background daemon running. To update, run npm install -g @openai/[email protected].

The command center, without the daemon

Run codex agents --no-daemon and the command center opens. The command center and the session picker now mark unread results, so finished work from background agents stands out instead of blending into the list. Tasks are sorted by recency within their status groups, so the most recent finished work is the first thing you see.

That matters because Codex sessions multiply fast. When several agents are running in parallel, the failure mode is not that the work doesn't get done — it's that you never notice it finished. The unread markers turn the command center into an inbox for agent output.

Switch backends, keep your drafts

The session picker can now switch between connected backends — and your drafts survive the switch. Saved chats also stay visible after you change model providers. If a chat can't resume on the new provider, you can still read its transcript and prepare an editable prompt to continue in a new chat, instead of losing the thread.

It is the same idea as the unread markers: session state should survive the moves you make between providers. Codex has been pushing toward longer, resumable agent sessions for a while; 0.167.0 closes a few of the gaps where that state used to fall through.

Images, patches, and Vertex AI

You can preview local image attachments while composing on supported terminals, and viewed local images open from clickable paths in the transcript. Patch details now expand directly in the transcript while an approval prompt is open, so you can read the diff before you approve it.

Sign-in gets a tidier path too: API key and cloud provider options sit under "Sign in another way," and enabling gcp_setup_wizard walks through Google Cloud Vertex AI setup with Application Default Credentials or a credential file. That is the second Google Cloud route for Codex in a month, after the model-garden listing in October. macOS and Windows users can also opt into system proxy settings from /experimental.

The smaller fixes

Ctrl+L now clears the terminal display while keeping earlier messages available through Ctrl+T. Network approval prompts close when their requests end, and concurrent requests to the same endpoint prompt one at a time instead of stacking. Long realtime V3 conversations save completed speech to chat history as the transcript grows, and automatic context compaction now accounts for retained output from failed responses and responses with missing token-usage data.

Nothing in 0.167.0 changes which model Codex runs or how it codes — the release notes carry no benchmarks or productivity claims. It is a workflow release for people already running several agents at once: the work is easier to find, harder to lose, and quicker to review. If you run Codex from the terminal daily, the unread markers alone are worth the update. OpenAI's recent moves around Codex suggest more of this session-management work is coming; for now, 0.167.0 is a solid, unglamorous step.