Codex CLI · 3 min read

Agent command center opens from the CLI, and the session picker marks unread results

Codex CLI 0.167.0 adds `codex agents --no-daemon` to open the agent command center, with unread markers in the command center and session picker. The release also brings backend switching with kept drafts, opt-in system proxy settings and expandable patches in approval prompts.

Key takeaways 5
  • `codex agents --no-daemon` opens the agent command center from the CLI, and the command center and session picker mark unread results.
  • The terminal session picker switches between connected backends and keeps your drafts.
  • `/experimental` has an opt-in setting for the macOS or Windows system proxy.
  • Approval prompts let you expand patch details in the transcript while the prompt is open.
  • Smaller changes cover Google Cloud Vertex AI setup behind `gcp_setup_wizard`, saved chats that survive provider switches, `Ctrl+L` clearing the display, tidier network approval prompts and better context compaction.

When several agents run at once, the hard part is knowing which one has something for you. Codex CLI 0.167.0 addresses that with a command center you can open from the terminal and with unread markers on finished results. Around it sit smaller fixes for switching backends, proxy setups and approval prompts.

This update covers November 6, 2026, and a single release, 0.167.0, with 14 changelog entries. Four are main features, the rest are minor wins.


Agent command center: see which agents have news

Open it from the CLI. Run codex agents --no-daemon and the agent command center opens straight from your shell. Check the documentation for what the --no-daemon flag changes in practice, because the changelog line only names it.

Unread results are marked. Both the command center and the session picker now flag results you have not looked at yet. If you keep a few agents working in parallel, you no longer have to open each session to find out whether it finished something. You scan the list and go to the marked ones.

Conditions. The changelog gives no plan or availability restrictions for this feature, and no prerequisites beyond version 0.167.0. If the command does not appear for you, check your version first.


Changes for everyday work

Switch backends from the session picker. The terminal session picker lets you move between connected backends, and your drafts stay put when you do. A half-written prompt used to be a reason to hesitate before switching. Now it is not.

System proxy settings, opt-in. On macOS and Windows you can turn on the system proxy settings from /experimental. Nothing changes until you enable it, so existing setups keep behaving as before. If you work behind a corporate proxy, this is the place to look.

Expand patches while approving. With an approval prompt open, you can expand patch details directly in the transcript. You read the change where it sits, then answer the prompt. That is a small change with a real effect on how carefully you can review.


Technical notes

Sign-in and providers. API key and cloud provider options now live under ‘Sign in another way’. If you want to set up Google Cloud Vertex AI, enable gcp_setup_wizard. It supports Application Default Credentials or a credential file.

Saved chats. Saved chats stay visible after you switch model providers. When a chat cannot resume, you can still read its transcript and prepare an editable prompt to continue in a new chat. The old conversation is not lost, even if it cannot be continued directly.

Terminal display. Ctrl+L clears the terminal display. Earlier messages remain available through Ctrl+T, so clearing the screen does not cost you the history.

Network approvals. Network approval prompts close when their requests end. Concurrent requests to the same endpoint now prompt one at a time, which should cut the pile of duplicate questions.

Realtime. Long realtime V3 conversations save completed speech to chat history as the transcript grows.

Context compaction. Automatic compaction now accounts for retained output from failed responses and from responses with missing token usage data. This changes how the context size is counted in those cases. The changelog does not say more, so watch your long sessions rather than assume a specific effect.


What to take away

If you work mainly in the terminal:

  • Try Ctrl+L followed by Ctrl+T to confirm that the cleared history is still reachable.
  • Expand a patch the next time an approval prompt appears and see whether it changes how you review.
  • If you use Vertex AI, enable gcp_setup_wizard and run through the setup.

If you run parallel sessions:

  • Open codex agents --no-daemon and check how the unread markers behave with your real workload.
  • Look at the same markers in the session picker.
  • Switch backends with a draft in progress and verify the draft is still there.

If you administer integrations or company deployments:

  • Test the system proxy option in /experimental on a macOS and a Windows machine. It is opt-in, so you decide who gets it.
  • Check the new ‘Sign in another way’ location for API key and cloud provider options, in case your internal instructions point to the old one.
  • Watch the network approval prompts, since concurrent requests to one endpoint now appear one at a time.

Sources: Codex CLI changelog, version 0.167.0 (November 6, 2026), and the Codex CLI documentation. The system proxy option sits under /experimental and the Vertex AI wizard behind the gcp_setup_wizard flag, so availability of both may change.

Patrick Zandl trains teams in Claude Code and agentic software development in Prague and writes about what actually works in real-world AI-assisted projects. Follow on LinkedIn

Read next

Get the next one by email

Every Tuesday: what changed in Claude Code, Codex and other coding agents, plus new articles. No spam.