Regression: #27801 removed TUI realtime voice support for API-key users

What version of Codex CLI is running?

Upstream main after #27801 (Remove TUI realtime voice support, merged 2026-06-12). This may affect the first released CLI version (likely 0.140.0) that includes that PR.

What subscription do you have?

Relevant to this issue is OpenAI API key auth.

Which model were you using?

Not model-specific.

What platform is your computer?

The removed TUI realtime/voice code path is platform-agnostic.

What terminal emulator and version are you using (if applicable)?

iTerm2, though this is not terminal-emulator-specific.

Codex doctor report

Not applicable. The regression is visible from the merged code removal in #27801 rather than from local environment state.

What issue are you seeing?

#27801 removed TUI realtime voice support for Codex CLI users authenticated with an OpenAI API key.

The PR removed the /realtime and realtime /settings TUI command paths, deleted voice capture/playback and WebRTC session handling, and removed the TUI-only realtime dependencies. That means the terminal-first voice/realtime workflow is no longer available in the CLI/TUI when using API-key auth.

This appears to regress the TUI voice transcription request tracked in #14630. That issue is still open and has multiple users asking for CLI/TUI voice dictation support.

What steps can reproduce the bug?

1. Use Codex CLI built from upstream main after #27801.
2. Authenticate with an OpenAI API key.
3. Start the TUI.
4. Try to use the previous TUI realtime/voice entry points such as /realtime or realtime audio settings.
5. The command path and implementation are gone.

What is the expected behavior?

TUI realtime voice/dictation should remain available for API-key users, or there should be a documented replacement/migration path that preserves a terminal-first voice input workflow.

Additional information

Related issues/comments:

• Original open feature request: #14630
• Removal PR: #27801
• Earlier related reports were closed as duplicates of #14630, so this regression can otherwise be easy to miss as only a comment on the older enhancement thread.