Feature request: Add native microphone and voice dictation to the Codex chat input

Summary

Please add an optional microphone / voice dictation control directly to the Codex chat input, especially in the IDE extension and follow-up workflow used after reviewing generated changes.

Problem

When working with Codex on real projects, follow-up instructions are often detailed and iterative: describe a failed test, clarify expected behavior, list several files to update, or explain a CI/CD scenario. Typing these instructions repeatedly slows down the feedback loop, particularly when the user already has the relevant code and diff open in the editor.

Today, users must rely on external speech-to-text tools, dictate into another application, and then copy the text into Codex. This breaks focus, adds friction, and often provides poor support for languages other than English.

Proposed experience

Add a microphone button near the Codex message composer and the follow-up changes input.

Possible behavior:

• Click or hold to record, then convert speech to editable text before submission.
• Support push-to-talk and an optional keyboard shortcut.
• Let the user review and edit the transcription before sending it.
• Support multilingual dictation and automatic language detection, including Ukrainian.
• Clearly indicate when recording is active.
• Keep the feature optional and disabled until the user activates it.
• Provide transparent privacy controls, including whether audio is processed locally or remotely and whether recordings are retained.

Why this would help

• Faster iteration when reviewing and requesting follow-up changes.
• Less context switching between the IDE, dictation software, and clipboard.
• Better accessibility for users who cannot comfortably type long prompts.
• More natural interaction for complex engineering explanations, debugging context, and CI/CD investigations.
• Stronger usability differentiation for Codex compared with coding assistants that only support keyboard input.
• Potentially higher product engagement because users can provide richer context with less effort.

Example workflow

1. Codex proposes changes and shows the diff.
2. The user presses the microphone button in the follow-up field.
3. The user says: “The implementation is mostly correct, but keep the existing pipeline parameter names, update only the Enlight validation stage, and add a fallback when the previous major version is unavailable.”
4. Codex transcribes the request into editable text.
5. The user corrects any technical names and submits it.

This would make iterative work with Codex considerably faster without changing the underlying agent workflow.