Skip to content

Codex CLI Voice Input Was Removed in 0.118.0: Use Desktop Dictation Instead

For / Key Points

For: Developers who followed older voice_transcription instructions but cannot use voice input in the current Codex CLI

Key Points:

  • Hold-Space voice transcription was removed from Codex CLI 0.118.0 on every OS
  • Linux and WSL audio capture was disabled even while the feature existed
  • Codex Desktop Global Dictation can enter text into a CLI prompt today

Hold-Space voice transcription is unavailable in the current Codex CLI on macOS, Windows, and Linux. PowerShell is not the cause, and the removal was not limited to Windows. OpenAI removed the unfinished implementation from CLI 0.118.0.

This article asks when and why voice transcription disappeared, and what current Codex users should use instead.

Conclusion: removed from CLI, supported on Desktop

As of July 15, 2026, the platform split looks like this.

EnvironmentHold-Space CLI transcriptionCurrent option
Windows PowerShell / TerminalUnavailableCodex Desktop Dictation or Win + H
macOS TerminalUnavailableCodex Desktop Dictation or macOS Dictation
Linux / WSLUnavailableHost OS dictation
Codex DesktopSeparate from the CLI featureDictation is supported

An OpenAI maintainer stated that voice transcription is fully supported in the desktop app but is not available in the TUI1. OpenAI has not published a product decision saying that the CLI feature was formally “migrated” to Desktop. The precise description is that the unfinished CLI path was removed while Desktop provides a supported alternative.

The feature lasted about five weeks

Voice transcription did not enter Codex CLI as a finished feature. Pull request #3381 added hold-Space recording and transcription on February 23, 2026. Its description called the feature “still in development” and required users to set features.voice_transcription = true2.

On February 27, an OpenAI maintainer responded to a WSL report by saying that the feature was not ready for use and would be marked experimental when ready3. The instructions spread before the feature had reached even that experimental stage.

On March 29, pull request #16114 removed the partially completed TUI transcription path. The deletion covered the feature flag, recording and transcription events, and the hold-to-talk state machine. That change shipped in Codex CLI 0.118.04.

2026-02-23  #3381   Hold-Space transcription added
2026-02-27          Maintainer says it is not ready for use
2026-03-29  #16114  Unfinished implementation removed

The published reason stops at incomplete and unsupported. Quality, maintenance cost, or a deliberate consolidation with Desktop may be plausible explanations, but OpenAI has not documented them as the reason.

WSL never worked, even before removal

Downgrading to 0.105.0 does not restore this feature on WSL. In that version, the Linux feature gate was disabled and the Linux recording handlers performed no action. Native Linux users reproduced the same behavior3.

When the Codex Windows app runs its agent in WSL, Codex CLI follows the Linux implementation. Adding voice_transcription = true to config.toml therefore did not start recording.

In current CLI 0.144.4, the flag itself is no longer recognized.

$ codex --enable voice_transcription features list
Error: Unknown feature flag: voice_transcription

The early implementation did work for some macOS users, but it is gone from current macOS builds too. A macOS user reported its removal in 0.118.0, and a separate request to restore TUI transcription remains open5.

realtime_conversation is a separate experiment

Enabling realtime_conversation does not restore Hold-Space dictation. It was a separate experiment for full-duplex audio through /realtime.

The flag remains in current builds, but the TUI's /realtime command, microphone capture, WebRTC path, and audio-device UI were removed on June 12, 20266. Experimental Realtime endpoints remain in App Server, so a visible flag is not evidence that the CLI has a usable voice interface.

Use Codex Desktop Global Dictation today

Global Dictation is available on the Windows PC used for this investigation. Codex Desktop 26.707.9981.0 exposed microphone selection, hold and toggle hotkeys, a custom dictionary, and transcription history. The official manual also documents Desktop Dictation and its shortcut7.

Global Dictation inserts text wherever the desktop cursor is focused. It can therefore target not only the Codex Desktop composer but also a Codex CLI prompt running in PowerShell or Windows Terminal.

  1. Open Settings → Voice → Dictation in Codex Desktop
  2. Select a microphone and configure a hold or toggle hotkey
  3. Focus the Codex prompt in the terminal and record with that hotkey
  4. Review the transcription before submitting it

If the hotkey does not respond

Global Dictation can conflict with another ChatGPT Desktop instance. A conflict appeared in the logs on this PC. Fully quit ChatGPT Desktop, restart Codex Desktop, and try again.

Without Codex Desktop, Windows voice typing through Win + H is another option8. Both approaches send ordinary text to the CLI and avoid direct microphone access from WSL.

Summary: the old article was version-specific

Instructions written for 0.105.0 accurately described code that existed at the time. However, that code never became supported and was removed from every OS in 0.118.0, so those instructions are now outdated.

The broader lesson is that an under development feature flag can change or disappear without becoming a supported product feature. A value left in config.toml does not prove that a current build can use it.

Today, use Codex Desktop Global Dictation first and keep OS dictation as a fallback. If you need native CLI transcription, follow open request #14630 and future release notes rather than relying on the old flag.

  1. OpenAI Codex GitHub Issue, TUI voice transcription. OpenAI's Eric Traut explains Desktop support and current TUI unavailability. ↩

  2. OpenAI Codex GitHub Pull Request, voice transcription, merged February 23, 2026. ↩

  3. OpenAI Codex GitHub Issue, Voice input not working on WSL. The thread contains the Linux code analysis and maintainer response. ↩↩

  4. OpenAI Codex GitHub Pull Request, Remove TUI voice transcription feature, merged March 29, 2026. ↩

  5. OpenAI Codex GitHub Issues, TUI voice transcription and Voice transcription for TUI. ↩

  6. OpenAI Codex GitHub Pull Request, Remove realtime conversation mode from TUI, merged June 12, 2026. ↩

  7. OpenAI, ChatGPT desktop app commands. ↩

  8. Microsoft Support, Use voice typing to talk instead of type on your PC. ↩