ChatGPT Voice Can Start and Steer Codex Tasks on Desktop, With Mobile Remote Access

OpenAI is documenting a hands-free workflow for starting and steering Codex with ChatGPT Voice. In an approved OpenAI Developers post on X, OpenAI says Voice can be used while building with Codex on desktop and mobile.

The durable ChatGPT Voice documentation fills in the workflow: Voice runs in the ChatGPT desktop app, can start or monitor Codex tasks, and can send follow-up instructions while work continues. OpenAI also documents mobile access through Remote on iOS, after pairing an iPhone with a desktop host.

This is not a separate mobile Codex runtime. The phone is a remote control and voice surface; the connected desktop host remains the machine running the work. That distinction matters for local files, permissions, and whether the host is awake and available.

Key Takeaways

• ChatGPT Voice can start, monitor, and steer Codex tasks from the ChatGPT desktop app.
• Voice can check progress, blockers, and results while a Codex task continues.
• Mobile access is through Remote on iOS after pairing an iPhone with a desktop host.
• The desktop host runs the work in the mobile workflow; the phone provides remote access.
• A chat or task must begin in voice mode to use Voice; a chat started in another mode offers voice dictation instead.
• Voice-driven tasks continue to use the Codex usage budget.

What Voice Adds to Codex

The normal Codex workflow assumes that you are looking at the desktop app, terminal, or editor while the agent works. Voice changes the interaction loop: you can describe a task aloud, ask for a status update, and change direction without returning to the keyboard. OpenAI’s documentation describes Voice as a way to coordinate tasks in Chat, Work, and Codex in the ChatGPT desktop app. (ChatGPT Voice documentation)

For Codex specifically, the documented actions include:

• Starting a separate thread for a longer task.
• Checking an existing thread.
• Sending follow-up instructions.
• Bringing progress, blockers, and results back into the voice conversation while work continues.

That makes Voice a supervisory layer over Codex rather than a replacement for the task environment. Codex still performs the work; Voice gives you a conversational way to launch it and keep it moving.

How to Start a Voice-Driven Codex Task

OpenAI’s documented setup is deliberately specific:

1. Open a new, empty chat or task in the ChatGPT desktop app.
2. Select Start new voice chat before sending a message.
3. The first time, allow microphone access, choose a voice, and review screen context on macOS.
4. Ask ChatGPT Voice to start a Codex task, such as running tests and investigating anything that fails.
5. Continue speaking to ask for progress, provide a follow-up instruction, or change direction.

A chat or task that starts in another mode supports voice dictation, but it does not become a full Voice session under the documented flow. To resume an earlier voice chat, open it and select Start voice chat. (ChatGPT Voice documentation)

Mobile Access Through Remote on iOS

OpenAI documents ChatGPT Voice on mobile through Remote on iOS after pairing an iPhone with a desktop host. In this arrangement, the phone is the place where you speak and monitor the work, while the paired desktop app provides the host environment for the Codex task. (ChatGPT Voice documentation; OpenAI remote connections documentation)

The host boundary is the practical caveat. A phone can be convenient for checking a task while away from the desk, but the desktop host still needs to be the machine that owns the project context and runs the work. Remote access does not turn a mobile device into an independent local Codex checkout.

For a reliable setup, pair the phone with the desktop host before you need it, keep the host available, and verify that the account and workspace have Codex access. If you need to work with local files or approve actions, expect those operations to remain tied to the host’s environment and permissions. (OpenAI remote connections documentation)

What to Say to Voice

Voice is most useful when the spoken request names the outcome and the handoff clearly. Examples:

• “Start a Codex task to run the tests and investigate anything that does not pass.”
• “Check the existing Codex thread and tell me what is blocking it.”
• “Send a follow-up asking Codex to update the tests, then tell me when it finishes.”
• “Start a separate thread for the migration plan and keep me posted on blockers.”

The first example comes directly from OpenAI’s Voice documentation. The other patterns follow the documented actions—checking threads, sending follow-ups, and bringing progress back to the voice conversation—without assuming extra automation or permissions. (ChatGPT Voice documentation)

Usage and Availability Boundaries

OpenAI says tasks started through Voice continue to use the Codex usage budget. Voice conversations also have their own plan-dependent allowance measured in rolling five-hour windows, so a voice-driven workflow has both the task’s Codex usage and the Voice session’s availability to consider. (ChatGPT Voice documentation)

The documented desktop plans are ChatGPT Plus, Pro, Business, Edu, and Enterprise. Enterprise and Edu availability begins with a two-week early-access period before the feature becomes available by default. Mobile access is documented through Remote on iOS, and availability can still depend on rollout status and workspace settings. (ChatGPT Voice documentation)

Only one voice chat can be active across the ChatGPT desktop app at a time. If Voice will not start, OpenAI recommends checking plan, rollout, and workspace availability, microphone permissions, and whether another voice chat is already active. (ChatGPT Voice documentation)

Bottom Line

ChatGPT Voice gives Codex a conversational control loop: start a task, check the thread, send follow-ups, and hear about progress without staying at the keyboard. The desktop app is the primary surface, while Remote on iOS lets a paired iPhone act as the mobile control surface for work running on the desktop host. Treat the host, workspace permissions, Codex budget, and Voice limits as part of the setup.

Sources

• OpenAI Developers on X, “Build hands-free using voice in Codex—on desktop and mobile” (August 23, 2026): https://x.com/OpenAIDevs/status/2091669454971343021
• OpenAI Developers, “ChatGPT Voice”: https://developers.openai.com/codex/features/voice
• OpenAI Developers, “Remote connections”: https://developers.openai.com/codex/remote-connections