--- title: "Codex CLI 0.163 Update: 6 New Features and Settings to Enable" description: "Codex CLI 0.163.0 shipped Oct 15 with mid-reply steering and a searchable /config. Here are the new features, an upgrade checklist, and Windows/WSL fixes." url: https://cdnping.org/en/blog/codex-cli-0-163-new-features language: en published: 2026-10-19 category: "Developer tools" tags: ["Codex","Codex CLI","Changelog","Developer tools","OpenAI"] source: cdnping.org license: Free to quote and summarize with a link to the url above. --- # Codex CLI 0.163 Update: 6 New Features and Settings to Enable Codex CLI 0.163.0 was released on October 15. Two changes matter most in day-to-day use: you can now send a follow-up message while Codex is still replying to steer it, and `/config` is now a searchable settings panel where some changes take effect immediately. The rest are a configurable leader key in `/keymap`, `/daemon`, parallel read-only tools, and Daybreak. This article is based on OpenAI's [official changelog](https://learn.chatgpt.com/docs/changelog), cross-checked against the third-party update aggregator [releasebot](https://releasebot.io/updates/openai/codex). The two sources disagree on the 0.163.0 date (October 15 in the official log, October 16 on the aggregator); we go with the official one. Everything here is accurate as of October 2026. ## What exactly changed in 0.163.0? | Change | Key points from the official description | | --- | --- | | `/config` panel | Searchable preferences; supported changes apply immediately | | Mid-reply steering | A follow-up message sent while Codex is replying steers it right away; an interrupt also stops any running Code Mode cell | | `/keymap` leader key | Configurable, default `Ctrl+X`; voice mute moved to `F9` | | `/daemon` | Restarts the managed background service | | Parallel read-only tools | Independent read-only tools can run at the same time | | Daybreak | On by default for eligible accounts; also available to `codex exec` with an OpenAI API key when the server offers it | The changelog does not explain what Daybreak does or who qualifies. Check your account page and the official documentation to see whether it is enabled for you. ## How do you steer Codex mid-reply? When Codex went off track, the usual move was to interrupt it and describe the task again. Now you can send the next message while it is still replying, and Codex adjusts immediately. This helps in cases like these: - It is editing a directory you don't want touched: add "don't change the tests directory." - The requirements picked up a new constraint, and you don't want to restart the task. - It is reading a lot of irrelevant files: tell it to look only at one module. Note that, per the official description, an interrupt also stops any running Code Mode cell. If you only want to add information, sending a message is gentler than hitting the interrupt key. ## Which /config settings are worth checking first? `/config` is now searchable, so you no longer have to dig through the docs for a setting name. Which options appear depends on your version and account, so trust the panel on your machine and the official docs. After upgrading, spend two minutes on these three things: 1. Run `/config` and search for the settings you change most often (approvals, sandbox, model-related) to confirm their current values. 2. Open `/keymap` and check whether the leader key (default `Ctrl+X`) collides with your terminal multiplexer's shortcuts, such as tmux. 3. If you use voice input, remember the mute key is now `F9`. ## How do you upgrade and confirm the version? Upgrade the same way you installed. For an npm install: ```bash npm install -g @openai/codex@latest codex --version ``` If you installed with Homebrew or the Windows installer, use the matching upgrade command from the official docs. You should see 0.163.0 or higher. After upgrading, open an old session: the official notes say resuming a thread restores its permissions and network approvals, so this is a quick way to verify. ## What else is worth catching up on from 0.162.0 and 0.161.0? According to the official log, 0.162.0 (October 8) added: - Tools to create and list managed Git worktrees (behind a feature flag), useful when you run several tasks in parallel. - Pinning tasks in Command Center with `p`. - `/copy` to copy conversation blocks, and `tui.mouse_scroll_speed` to adjust scrolling speed. - Respecting the server's `Retry-After` hint on retryable failures, and `apply_patch` now preserves CRLF line endings. The 0.161.0 contents (terminal login with `/mcp login `, voice input/output device selection, and so on) have slightly different dates across sources, so check the official log. Also, on October 9 the Codex desktop app launched Composer predictions (beta): after a reply, it suggests a follow-up in the input box, and pressing Tab accepts it as an editable draft that is not sent automatically. The official notes say it is limited to individual ChatGPT Pro users aged 18 or older in supported regions. ## What was fixed for Windows and WSL users? In 0.163.0: - On Windows, `Ctrl+V` and `Shift+Insert` now paste both images and text. - On WSL, repeated pastes reuse the clipboard reader, so a copy timeout no longer hangs it. 0.162.0 also restored file access on plain drive letters on Windows 10 and fixed an archive verification problem when installing through Windows PowerShell. ## How do you tell what's wrong after an upgrade error? First separate local configuration from server-side problems. For server capacity or outages, check cdnping's [Is Codex down?](/en/down/codex) and [OpenAI API status](/en/status/openai) pages. For October incident records, see [Codex errors and "Selected model is at capacity" troubleshooting](/en/blog/codex-outages-october-2026-capacity). If it looks like a network or proxy problem, see [Claude Code vs Codex: domains, ports, and proxy settings](/en/blog/claude-code-vs-codex-network-requirements). For model-level changes, see [Codex retires its old default model on October 14](/en/blog/codex-model-retirement-cli-0-163) and [GPT-6.1 Sol becomes the Codex default model](/en/blog/gpt-6-1-sol-codex-default). ## Can you point Codex at a different endpoint? Codex supports a custom API endpoint; the setup is covered in [Custom API endpoints for Claude Code, Codex, and OpenCode](/en/blog/configure-custom-api-endpoint-claude-code-codex). Disclosure: AGI Plan is a commercial partner of cdnping (a commercial partner). It provides a compatible endpoint with a step-by-step Codex tutorial ([start here](https://agiplan.dev/startup)), billed by monthly subscription and settled in RMB (Chinese yuan). It is not an officially authorized channel. Whether it suits you depends on your requirements for model sourcing and data policy, so review its documentation yourself before using it. ## Sources - [OpenAI Codex changelog](https://learn.chatgpt.com/docs/changelog) (0.163.0, 0.162.0, Composer predictions) - [Releasebot: OpenAI Codex updates](https://releasebot.io/updates/openai/codex) (0.161.0 and Windows/WSL items, cross-checked)