Codex CLI 0.171.0: the TUI workspace sidebar is on by default, GPT-6.2 Sol is the default model, and codex exec gets --remote
Codex CLI 0.171.0, now the npm latest, turns on the TUI workspace sidebar by default, makes GPT-6.2 Sol the default catalog model, and adds codex exec --remote, /rc remote control and undo/redo in the prompt composer. What changes when you upgrade, and the commands involved.
OpenAI released Codex CLI 0.171.0 (tag rust-v0.171.0 in the openai/codex GitHub repo) on November 20, 2026, two days after 0.170.0, and it is now the latest tag on npm. Four changes show up in the terminal UI: a workspace sidebar is on by default in the terminal UI, GPT-6.2 Sol is the default model in the catalog, the prompt box has undo and redo, and /rc manages remote control. For scripts, codex exec can now run a turn on an existing app server with --remote.
Install or upgrade with the command from OpenAI's changelog:
npm install -g @openai/[email protected]What changes when you upgrade
- TUI workspace on by default: Sidebar tabs for file previews, image previews and subagent transcripts next to the chat (PR #61352 and follow-ups).
- GPT-6.2 Sol default: gpt-6.2-sol is the default catalog selection unless your config or --model says otherwise (PR #62106).
- Composer undo/redo: Alt+Z (Unix), Ctrl+Z (Windows) or F7 undoes; add Shift to redo (PR #61954).
- codex exec --remote <ADDR>: Non-interactive turns on a running app server, new or resumed thread (PR #61707).
- /rc and /remote: Turn remote control on or off and show a pairing code from the TUI (PR #62121).
- /statusline providers: Pick or write an MCP-backed status line, behind tui_extensions_statusline (Release notes).
The workspace sidebar
The workspace had been an under-development feature. PR #61352 moves the tui_workspace feature to stable and sets it to enabled by default, which is why it shows up after the upgrade without any config change.
What lives in it, according to the merged pull requests:
- File previews: file links in the transcript open as sidebar tabs and render as Markdown. You can search, select and copy text, move between links with Tab and Shift+Tab, open one with Enter, zoom, and close the tab with Esc.
- Image previews: PNG, JPEG, GIF and WebP links open in the sidebar, sized to fit, with format and dimension labels. When the terminal cannot draw graphics you get a fallback instead.
- Subagents: the sidebar chooser has a Subagents section with each agent's name and status (Completed, Stopped and Failed stay visible). Selecting one opens its transcript in a read-only tab beside the parent conversation instead of switching away from it.
If you prefer the old single-pane view, OpenAI has not published a switch for it. Because the setting is an ordinary feature flag named tui_workspace, the general codex features command should handle it: run codex features list to confirm the name and its state, then codex features disable tui_workspace, which the CLI reference says writes the change to $CODEX_HOME/config.toml. codex features enable tui_workspace puts it back. This is our reading of the flag, not a documented procedure.
codex features list
codex features disable tui_workspace # back to the single-pane TUI
codex features enable tui_workspace # sidebar againGPT-6.2 Sol becomes the default
PR #62106 adds gpt-6.2-sol and gpt-6.2-astra to the bundled, Amazon Bedrock and Vertex model catalogs and makes GPT-6.2 Sol the default selection. The same change raises the GPT-6.1 Sol context window in the catalog to 500,000 tokens and lists an ultrafast service tier for GPT-6.2 Astra on Bedrock. If you pin a model in config.toml or pass --model, that still wins; the CLI reference uses --model gpt-6.2-sol as its example.
Undo and redo in the prompt box
The non-Vim composer now keeps an edit history. Default undo keys are Alt+Z on Unix, Ctrl+Z on Windows and F7, with Super+Z also available on macOS; hold Shift with the same key to redo. Typing runs and Backspace runs are grouped into separate steps, pastes are kept as their own step, and undo restores whole drafts including attachments and large pastes. The actions are named composer.undo and composer.redo in the keymap picker if you want different keys. Vim mode is not covered by this change.
codex exec --remote for scripts and CI
codex exec is the non-interactive mode used for scripts and CI. With 0.171.0 it accepts --remote <ADDR> and runs the turn on an app server that is already running, using that server's configuration and credentials rather than the local ones. PR #61707 says it can start a new thread or resume one by UUID, prints human-readable or JSON output, and takes a bearer token through --remote-auth-token-env. It refuses conflicting endpoints, agent configuration overrides it cannot apply, and threads that are already active, and it reports an error if the server disconnects before the turn finishes.
The address formats are the ones the CLI reference lists for the global --remote flag: ws://host:port, wss://host:port, unix:// for the default socket, or unix://PATH. The server side is codex app-server --listen. The reference notes that the app server itself accepts ws:// listen URLs and that you need TLS termination or a proxy for wss://, and that a token is only sent over wss:// or local-only ws:// URLs. A minimal local setup looks like this; the prompt is a placeholder:
# terminal 1: start an app server on the default Unix socket
codex app-server --listen unix://
# terminal 2: run one non-interactive turn against it
codex exec --remote unix:// "[your task]"At the time of writing, the CLI reference's --remote entry does not list codex exec among the supported subcommands (it says other subcommands reject remote mode), so the release notes and PR #61707 are the only documentation. PR #61707 also says the remote options can go before or after exec.
/rc, /remote and the status line
/rc (alias /remote) opens a menu that turns remote control on or off for the managed local background server, tells you whether the preference was saved for the next restart, and, once enabled, shows a manual pairing code under Add device. It only affects local sessions, even if you open it while viewing a cloud session. From the shell, the related commands are codex remote-control start, stop and pair; pair --json returns pairingCode, manualPairingCode, environmentId and expiresAt.
/statusline already let you pick and reorder footer items such as model, context, rate limits and git branch, saved to tui.status_line in config.toml. When the tui_extensions_statusline feature is enabled, it now also offers Choose a status line provider, which reads from MCP servers, and Create a custom status line, which opens a new conversation with a draft bound to the bundled create-statusline skill. The create option is hidden in cloud mode.
Known open issue: GPT-6.2 Sol cache gap
Now that GPT-6.2 Sol is the default, one open bug report is worth watching if you pay per token. Issue #62246 in openai/codex, filed on November 18 against Codex CLI 0.169.1 with a custom Responses-compatible provider, reports that on GPT-6.2 Sol the next request's cached input stops at the previous request's input and misses the previous output. In the reporter's measured pairs the gap was 309 to 865 tokens per request on GPT-6.2 Sol, against 161 to 178 on GPT-6.1 Sol, and it grows with the size of the previous output. The 6.1 numbers come from older CLI versions (0.164.0 to 0.167.2), so the reporter says it is not a perfectly matched comparison. The issue is open and labelled bug; the only reply so far is a +1 from another user, and nothing in the 0.171.0 notes mentions a fix. We have not reproduced it.
Fixes worth knowing
- Permission profiles are no longer lost when you change working directories.
- A late completion event can no longer cancel the approval prompt for a newer turn.
- Drafts, queued follow-ups and interrupted input stay with their original thread when you switch or resume sessions.
- Scrolling no longer inserts stray characters, and the caret no longer jumps while typing.
- Enterprise MCP sign-in works with identity providers that issue ID tokens without refresh tokens.
Try it with your AI. Give this to your coding agent to upgrade Codex CLI and set it up the way you want, step by step.
Help me upgrade OpenAI Codex CLI to 0.171.0 and check what changed for my setup. Show me
each command before running it and wait for my OK.
1. Run codex --version and npm view @openai/codex dist-tags. If the installed version is
already 0.171.0 or newer, skip step 2.
2. Upgrade with: npm install -g @openai/[email protected]. Do not use any other installer.
3. Run codex features list and tell me the state of tui_workspace and
tui_extensions_statusline. I want the workspace sidebar [on/off]. If off, propose
codex features disable tui_workspace and wait.
4. Read my $CODEX_HOME/config.toml (do not print secrets or tokens) and tell me whether
a model is pinned. If not, new sessions will use gpt-6.2-sol. I want [model name or
'the default'].
5. Optional: I want to run non-interactive jobs on a local app server. Propose starting
codex app-server --listen unix:// and running codex exec --remote unix:// "[test
task]" in a throwaway repo. Do not expose the server on a network address.
Do not delete files, do not edit credential files, and stop after reporting the result.FAQ
How do I install Codex CLI 0.171.0? Run npm install -g @openai/[email protected]. It is the latest tag on npm as of November 20, 2026; the npm alpha tag currently points to 0.173.0-alpha.1.
How do I turn off the new sidebar in Codex CLI? OpenAI has not documented a toggle, but the feature flag is tui_workspace, so codex features disable tui_workspace should restore the single-pane view. Check the flag name with codex features list first.
What is the default model in Codex CLI 0.171.0? GPT-6.2 Sol (gpt-6.2-sol). A model set in config.toml or passed with --model overrides it.
What does codex exec --remote do? It runs a non-interactive turn on an existing Codex app server, using that server's configuration and credentials, and can start a new thread or resume one by UUID.
What is the undo shortcut in the Codex prompt box? Alt+Z on Linux and macOS terminals, Ctrl+Z on Windows, F7 on any platform, and Super+Z on macOS. Add Shift to redo.
Which Codex commands accept --remote? The CLI reference lists codex, codex agents, queue, resume, fork, archive, delete and unarchive. Codex CLI 0.171.0 adds codex exec --remote, which the reference does not list yet.
Does /rc affect cloud sessions? No. Its actions apply only to local sessions, even when you open it while viewing a cloud session.
Which images can the Codex sidebar preview? PNG, JPEG, GIF and WebP, with a fallback when the terminal cannot draw graphics.
Related on claudefox
- GPT-6.2 Astra and Sol in the API: prices, the Ultrafast change in March, and what to fix before you switch
- GPT-6.2 benchmarks vs Claude Fable 5.2: what independent evaluators found
- tstack: nine agent skills that take a change to a merged PR in Claude Code and Codex
- Floorplan: a Claude Code and Codex plugin that maps your codebase in plain English
Source: openai/codex release rust-v0.171.0, Codex changelog (developers.openai.com), Codex CLI command line reference, PR #61352: Mark the TUI workspace as stable and enable it by default, PR #61707: Support remote app servers in codex exec, PR #62121: Add /rc to manage remote control from the TUI, PR #61954: Add undo and redo to the non-Vim prompt composer, PR #62106: Add GPT-6.2 Sol and Astra to model catalogs, @openai/codex on npm, Issue #62246: GPT-6.2 Sol implicit cache boundary excludes the previous assistant output
This story first appeared in the claudefox daily briefing for November 20, 2026.