• AI with ALLIE
  • Posts
  • Step-by-step instructions to set up voice mode in ChatGPT and Codex

Step-by-step instructions to set up voice mode in ChatGPT and Codex

Ten minutes to set up, and you can stop typing entirely

AI with ALLIE

The professional’s guide to quick AI bites for your personal life, work life, and beyond.

Today I'm walking you through the exact setup for Codex live voice mode, because it's an unreal feature that I can't stop talking about, and because most of you are a 10-minute setup away from doing your email on a walk instead of at your desk.

I want you talking to your computer by this time tomorrow. And if you're still getting your AI foundations in place first, my AI-First Academy was built for exactly that.

What is live voice mode?

It’s a thing that lets you yap while you get valuable agentic power out of AI.

Codex is OpenAI's agent. It can open the files sitting on your computer, run things in your browser, edit your documents, take actions inside your tools (like messaging in Slack, or emailing 100 custom emails in Gmail, or edit your Meta ads), and spin off other agents to work in parallel while you're off doing something else. ChatGPT is OpenAI's chat assistant, good for brainstorming, writing, and one-off questions. Codex is “part of” ChatGPT, according to OpenAI. Both Codex and the standard ChatGPT have live voice mode, and the one we really care about is the Codex one.

Codex live voice mode is a spoken conversation with an agent. You talk, it does work, and you can interrupt it mid-sentence with more thoughts or tasks. Underneath the feature is a new model called GPT-Live. GPT-Live is OpenAI's duplex speech model, and duplex just means you can both talk at once, the way a phone call works. AI is processing your speech and generating its own speech at the same time. Simultaneous awesomeness.

ChatGPT has a live voice mode too, running that same GPT-Live. It will happily chat with you, but it can't open your Chrome, read the ad on your screen, or fire off four agents while you keep talking. Honestly, I think it’s pretty dumb. Please, if you’re going to read this newsletter, use voice mode in Codex.

Voice mode removes AI prompting friction

Live voice mode removes the effort of prompting. And it means that instead of keeping a notepad next to me where I’m constantly jotting down random little tasks (email this person, track down this invoice, double check that contract), I just say it out loud to an AI chief of staff that is already waiting and listening for me. It takes AI from a one-off task rabbit to a persistent ambient workforce.

Voice mode makes it ridiculously easy to use AI for everyday tasks. Here is a video of me using it to prep for my AI workshop with Mark Cuban. This feature is so powerful, and you can be in the first 1% of AI users jumping on this.

It’s like having an extra teammate with 6 eyeballs.

The complete beginner walk-through (even if you’ve never used Codex)

Your computer is about to turn into a teammate and your phone is about to become a voice-controlled remote control for your Mac. The whole thing takes about ten minutes, even if you've never opened Codex in your life.

Step 1: Confirm you have a paid ChatGPT account

  • Sign in at chatgpt.com and check your plan under profile/settings.

  • For the complete workflow below (including Codex, mobile remote mode, and live voice mode), you have to be on a paid account (Plus, Pro, Business, Edu, or Enterprise).

  • If you use a company or school workspace, your admin may need to enable Codex, Remote Control, or Voice. Go ask nicely :)

Check that your account is: Plus, Pro, Business, Edu, or Enterprise by going to ‘profile/settings’.

Yay! You have the right account and access, and now we need to get the tool.

Step 2: Install the ChatGPT desktop app

  • On your actual computer, go to chatgpt.com/download.

  • Choose the macOS download (there is also a PC version that I'm sure follows the same steps, but I only have a Mac).

  • Open the downloaded installer and follow the onscreen installation steps.

  • Open the ChatGPT app from Applications (or hit command+space and search for 'ChatGPT' in Finder).

  • Sign in with the ChatGPT account and workspace you plan to use on your phone.

  • In the app, open the top-left product dropdown and select Codex.

  • Select New chat. If you want Codex to work with local files (aka the files that just live in your computer), open the relevant folder or create a project and approve the requested file permissions.

  • Install any available app update before continuing.

Open the top-left product dropdown, switch from ChatGPT and select Codex, and hit New chat.

Yay! You can do cool agent stuff on your computer, and now we want to get it on your phone too. Note: if you don't care about voice mode on the go, skip to Section 6.

Step 3: Install ChatGPT app on your phone

You’ve probably already done this, but just in case you haven’t:

  • On iPhone, open the App Store. On Android, open Google Play.

  • Search for ChatGPT and verify that the publisher is OpenAI.

  • Install or update the app to the latest version.

  • Sign in with the same ChatGPT account and workspace used on the Mac.

  • Complete any required multi-factor authentication, SSO, or passkey setup.

  • Sign in. Awesome.

In the App Store or Google Play, search ChatGPT, check the publisher says OpenAI, and sign in.

Yay! You have the right account on your laptop and phone, and now we have to make them talk to each other.

Step 4: Pair your phone with your desktop app for Remote Control

  • Leave the ChatGPT desktop app open on your computer.

  • On your computer in the desktop app, click your name in the bottom left, then go to Settings > Connections > Control this Mac or PC.

  • Select Set up or Add.

  • Approve remote access and complete any verification requested by the app.

  • A QR code will appear on your computer screen. Open up the camera app on your phone and "scan it" aka hold your phone up to the screen.

  • The QR code should open ChatGPT on the phone. Confirm that both devices show the same account and workspace, then finish the security checks it asks you about.

  • You'll know the pairing is successful when you go into the ChatGPT mobile app and on the left-hand side there is a new option called Remote. See screenshot below.

  • Back on your computer, on the same settings page (Settings > Connections), toggle ON 'Keep this Mac awake' so you can keep working remote without your computer shutting off. See screenshot below.

In the desktop app, click your name in the bottom left, then Settings > Connections > Control this Mac or PC.

A QR code appears and you hold your phone camera up to your monitor like you're reading a restaurant menu. You'll know it worked when a new option called Remote shows up on the left side of the mobile app.

Remote should now appear in your left-hand navigation bar

Under settings>connections, toggle on 'Keep this Mac awake.'

Security note: Remote is using your literal computer files, your signed-in websites, your credentials, plugins, and permissions. Only pair a phone you trust. Remove the device under Settings > Connections the second that phone is lost.

Yay! Everything talks to each other. Now it's time to yap to Codex and make our computer do stuff.

Step 5: Start using live voice mode from your phone

I use this method several times a week. It’s more for “I want to or need to be away from my computer but still have to get work out” vibes.

  • Confirm that your computer is awake, online, and running the ChatGPT desktop app.

  • On your iPhone, open ChatGPT.

  • Click Remote on the left and choose the paired Mac (it'll likely be the only option).

  • You may have to start a new, empty chat or task.

  • Before typing or sending anything, tap the blue waveform/Voice button at the bottom (see screenshot below).

  • The first time, allow microphone access and choose a voice (I use Cove).

  • Start talking. Literally just say hello. You can ask ChatGPT to begin work, check another task, report progress, run through your email, update a spreadsheet, research a production company, or change direction while you keep talking. More use cases are in Section 7.

  • Tap End when finished.

Tap the blue waveform button at the bottom. And literally just start with hello.

Important: the chat must begin in Voice mode. If you start the chat by typing first, you can only dictate, but you can't do live voice mode. In that case, you'll have to open a fresh chat and tap the Voice button again before sending the first message.

Step 6: Or, start using live voice mode from your laptop (no phone needed)

I use this method several times a week. It’s more for “oh man I’m sitting at my desk and have so much to get done, I want to be able to live pair with someone” vibes.

  • Open up the ChatGPT app on your laptop (again, you can search for it in Finder by pressing command+space and searching for 'ChatGPT').

  • Click New Chat in the top-left corner.

  • Click on the waveform button in the bottom right of the prompt window. See screenshot below.

Yay! Now you're ready!

Step 7: Genuinely useful ways to use voice AI

  • Morning rundown: “Get an agent to review my calendar, email, and messages, then walk me through today’s priorities, preparation I owe, deadlines, and open time.” Connected services and permissions need to be set up first. Or it'll give you a popup asking for access that you can approve (and annoyingly won't tell you that there's a popup, so you may have to hunt).

  • Email triage, one message at a time: "Fire off an agent and take me through my inbox one email at a time. Summarize each message, tell me whether it needs a reply, suggest a draft, wait for my feedback, then finalize the reply. Do not send, archive, or delete anything unless I explicitly approve it.”

  • Meeting preparation: “Fire off an agent to look at my next meeting, gather the relevant calendar notes, email history, and files, then give me a two-minute spoken briefing and three questions I should ask.”

  • Walk-and-talk work: Talk through a messy problem while Voice turns it into a plan, starts separate research or Codex tasks, and reports the results back to you.

  • Remote task management: Start a long task on the Mac, leave your desk, then use the phone to check progress, answer questions, approve an action, or redirect the work. Great if you want to walk away for lunch.

  • Work from what is on screen: On the Mac app, go to Settings > Voice > Screen context and toggle it on. Then say, “Take a look at this,” to discuss the frontmost window (careful with what data is on the screen). Ask it to weigh in on a presentation, ad, I ask it work work in one tab while I'm in another.

  • File and document help: Ask it to summarize a PDF, clean a spreadsheet, review a proposal, draft 18 follow-ups, review 20 invoices, edit 3 schedules, or create 5 checklists from a folder of materials.

  • Recurring briefs: Ask it to schedule a daily or weekly briefing that reviews your connected calendar, messages, and priorities.

  • Errands help: I was walking to Warby Parker and needed to reorder a previous pair that I lost. So I told Codex Live Voice Mode to jump on my computer, use chrome, access my personal email (which is not connected to Codex), search Warby in my email, find the recent orders, find the gold pair, and tell me what to order. I walked in and knew I needed a pair of "Gus" glasses without needing to look at my phone at all.

  • ANYTHING: literally anything. I asked it to walk me through the latest research in ADHD for 40min. I told it where I was walking in NYC and had it find two stores along the way that I might like (and direct me to them). I had it build me a tool to better manage invoices when I got back to my computer. You can have it control your Claude Code on your desktop. And you can have it do all of these in parallel.

STEAL THIS CODEX VOICE PROMPT

"Fire off an agent and take me through my inbox one email at a time. Summarize each message, tell me whether it needs a reply, suggest a draft, wait for my feedback, then finalize the reply. Do not send, archive, or delete anything unless I explicitly approve it."

It has a few issues, and it’s worth the tradeoff

Once you start using voice with AI, it completely changes how you work. Speaking is 4x faster than typing, it’s more natural, and it’s easier to get through multiple, even unrelated, tasks. But, like all new AI features, live voice mode still has its issues:

  • Your phone will heat up and your phone battery might run out more quickly.

  • The voice session only lasts 90min. When it ends, start a new one.

  • If Codex says it's "checking" something and then goes quiet, look at your phone screen. There might be a permission popup blocking everything, and the app will not tell you it's there.

  • I'm on the $200/mo plan and haven't hit my limits, but if you're going to really use this, I think you want at least the $100/mo plan or a team plan.

  • Always verify outputs and ask it to spin up a second agent to thoroughly double check the work of the first as if it will be graded on it.

My business recommendation: setup Codex live voice mode and Remote control today (or on a personal account if your work doesn’t allow for it), then take one walk tomorrow where you yap through your email. If any step loses you, copy this whole newsletter, drop it into Codex, and type "wtf? explain".

And don't clean up how you talk to it. Don’t edit yourself. Say exactly the thing you want to say. Say the thing you need.

Voice AI is the interface of the future. Yapping is the new prompting. Grab a gooseneck mic and some cough drops and go talk to your computer.

Stay curious, stay informed,

Allie

P.S. Claude does not have this feature. So if you want to use it, jump into the Codex ocean. The water’s warm.

UPSKILL WITH THE AI-FIRST ACADEMY

Sample moment from the AI-First Academy. Plus, interactive learning, grab-and-go prompts, and more!

If you want to go from AI foundations (I’m talkin’ total beginner!) to building your own agent workforce in less than 6 hours, with all the workflows I teach my Fortune 500 clients, join my AI-First Academy.

No coding required. The Academy is built for business people, not engineers. And everything is customized just for you - track your progress, engage with knowledge checks, and graduate the levels. We even have discounted team seats, for all the groups looking to level up together.

Feedback is a Gift

I would love to know what you thought of this newsletter and any feedback you have for me. Do you have a favorite part? Wish I would change something? Felt confused? Please reply and share your thoughts or just take the poll below so I can continue to improve and deliver value for you all.

What did you think of this month's newsletter?

Login or Subscribe to participate in polls.

About Allie

Learn more AI through Allie’s socials: LinkedIn, Instagram, X, TikTok

Want to work with Allie? Reach out here