OpenAI shipped Codex CLI 0.167.0 on 2026-11-06 (tag rust-v0.167.0, npm @openai/codex@0.167.0). Per the official release notes, multi-session management lands as Agent Command Center, with draft-keeping backend switches, local image preview, and a clearer alternate sign-in entry.
Takeaways:
- Open Agent Command Center with
codex agents --no-daemon; unread results are marked in the command center and session picker (per official release notes). - Switching connected backends from the terminal session picker keeps your drafts (per official release notes).
- On supported terminals you can preview local image attachments while composing, and open viewed local images from clickable transcript paths (per official release notes).
- API Key and cloud-provider options live under “Sign in another way”; enable
gcp_setup_wizardfor Google Cloud Vertex AI (per official release notes). - Hands-on with our API Key + custom provider: after upgrading to
codex-cli 0.167.0, a short file-edit task finished in about 22 seconds, and a one-line Q&A returned in about 3 seconds.
0.167.0 at a glance
| Change | Release-note gist | Who cares |
|---|---|---|
| Agent Command Center | codex agents --no-daemon; unread results in center + picker | Anyone juggling multiple Codex sessions |
| Multi-backend switch keeps drafts | Switch connected backends in the session picker without losing drafts | Multi-backend / multi-provider workflows |
| Local image preview | Preview local images while composing; clickable transcript paths | People attaching screenshots or diagrams |
| Sign in another way / Vertex | API Key + cloud options grouped there; gcp_setup_wizard for Vertex ADC or a credential file | Key-based or GCP sign-in instead of ChatGPT login |
| System proxy (experimental) | Opt into macOS / Windows system proxy from /experimental | Desktops that must use the OS proxy |
| Expand patch during approval | Expand patch details in the transcript while an approval prompt is open | Reviewers who read diffs before approving |
| History across providers | Saved chats stay visible after switching providers; if resume fails, read the transcript and prepare an editable prompt for a new chat | Frequent provider switchers |
Other fixes (all per official release notes): Ctrl+L clears the display while earlier messages stay reachable via Ctrl+T; network approval prompts close when requests end; concurrent same-endpoint approvals are serialized; automatic context compaction accounts for retained output from failed responses.
Version and prerequisites
codex --version
# Expect codex-cli 0.167.0; upgrade if lower
npm i -g @openai/codex@0.167.0
Our run showed:
codex-cli 0.167.0
New installs: Codex install guide. Broader usage: How to use Codex.
We used a global ~/.codex/config.toml custom provider matching Codex config.toml (base_url includes /chatgpt/v1; put the Key in experimental_bearer_token, not an env-bound Key field):
model = "gpt-6.1-sol"
model_reasoning_effort = "medium"
model_provider = "aicodewith"
[model_providers.aicodewith]
name = "aicodewith"
base_url = "https://api.aicodewith.ai/chatgpt/v1"
wire_api = "responses"
experimental_bearer_token = "YOUR_API_KEY"
For 401s, see Codex 401 Unauthorized and Codex API Key config.
Agent Command Center: how to open it
Per the official release notes, 0.167 folds multi-session browse/switch into Agent Command Center:
codex agents --no-daemon
--no-daemon runs the overview with an embedded app server. codex agents --help describes the same entry point. Unread results are marked in the center and picker (release notes #58263, #58264). We confirmed the codex agents subcommand ships with 0.167.0; the full interactive UI needs a real TTY—open it in your local terminal to see unread badges and the session list.
Multi-backend, local images, alternate sign-in (per official notes)
We describe these from the release notes and do not claim UI clicks we did not complete in this environment:
- Draft-keeping multi-backend switch: switching connected backends in the terminal session picker keeps drafts (#58280, #58457).
- Local image preview: preview local attachments while composing; open viewed paths from the transcript (#58261, #58446).
- Sign in another way: API Key and cloud options are grouped here;
gcp_setup_wizardsets up Vertex with ADC or a credential file (#57857, #58427). - System proxy: opt into macOS or Windows system proxy under
/experimental(#58157). - Expand patches during approval: expand patch details in the transcript while the approval prompt is open (#58281).
- History after provider changes: saved chats remain visible; if a chat cannot resume, you can still read its transcript and prepare an editable prompt for a new chat (#58202).
Hands-on: upgrade + short coding task
Environment: Linux, npm global @openai/codex@0.167.0, provider aicodewith, model gpt-6.1-sol, approval_policy=never, sandbox_mode=workspace-write.
- Upgrade: after
npm i -g @openai/codex@0.167.0,codex --versionprintscodex-cli 0.167.0. - Config: write
~/.codex/config.tomlas above (name = "aicodewith"+experimental_bearer_token+base_url = "https://api.aicodewith.ai/chatgpt/v1"). - Short coding task: temporary
hello.pywithadd; ask Codex to addmuland printmul(3,4)in__main__. About 22 seconds;python3 hello.pyprinted5and12. - Short Q&A: “only answer 78” returned `56` in about 3 seconds*.
Non-interactive example:
codex exec --skip-git-repo-check \
-c 'approval_policy="never"' \
-c 'sandbox_mode="workspace-write"' \
"In hello.py, add mul(a,b); print mul(3,4) in __main__."
For the interactive overview, use codex agents --no-daemon.
Remove the Key from config.toml when you finish testing.
FAQ
Do I have to reinstall for 0.167?
Only if you are below 0.167.0. Upgrade with npm i -g @openai/codex@0.167.0, or use codex update for your install channel.
Agent Command Center vs codex resume?
codex resume continues one session; codex agents is the overview/switch entry (optionally --no-daemon). Unread badges and cross-backend switching are 0.167 overview-side additions (per official release notes).
Do custom providers still work?
Yes. Keep the Key in global experimental_bearer_token, set name = "aicodewith", and use base_url = "https://api.aicodewith.ai/chatgpt/v1". Our short tasks used that setup. For 401s, see Codex 401 Unauthorized.
How do I enable the Windows / macOS system proxy?
Per the official release notes, choose it under /experimental. This hands-on run was on Linux, so we did not toggle that switch here.
Wrap-up
Codex 0.167.0 centers multi-session work in Agent Command Center and adds draft-keeping backend switches, local image preview, and clearer alternate sign-in (including the Vertex wizard). Upgrade to 0.167.0, write config.toml, open codex agents --no-daemon, then run tasks via codex exec or an interactive session. Our short coding task took about 22 seconds; the short Q&A about 3 seconds.
Related: How to use Codex · Install · config.toml · 401 · API Key config



