How to use voice input in Codex
Speak quickly and accurately with Flow—on Mac or your phone. Look things up while you keep recording.
- ~180 characters/min
- Reference pace for Chinese dictation
- Your vocabulary
- A personal dictionary for more accurate recognition
- Keep moving
- Look things up without interrupting your recording
How do you use Flow to dictate in Codex?
With Flow installed and set up, focus the Codex prompt field on Mac. With Flow's default shortcuts, hold Fn to speak and release to finish, or use Fn + Space for hands-free input. Flow recognizes and organizes your speech into text that you can review before sending.
- Focus the Codex prompt field. Open Codex on your computer, or your Codex conversation in ChatGPT on iPhone, with Flow set up for input.
- Start Flow and speak naturally. Explain the task, add context or give feedback. On desktop, use hands-free mode for a longer explanation. You can check other content while you speak.
- Review the text and continue the conversation. Flow recognizes and organizes your speech. Edit or add to it before sending, then use Flow for your next response to Codex.
Set up macOS or Windows shortcuts · Read more about speaking to AI
What if product names or specialist terms are misrecognized?
A misrecognized product name can send an AI down the wrong path. Correcting the same specialist term over and over eats into the time voice input saves. Flow focuses on recognizing what you say accurately, then making it clear to read.
A personal dictionary for the words you use at work
Add your product names, team abbreviations and specialist vocabulary to Flow's dictionary to improve recognition of those terms. Your input tool can adapt to your vocabulary, rather than relying only on common words.
Open Dictionary in Flow and select + to add a term, or import an existing word list in bulk. You can keep using that dictionary when explaining a task in Codex, exploring an idea in ChatGPT or writing to a colleague. See dictionary setup and sync help.
Room to finish a thought
Flow supports recordings up to 10 minutes and offers hands-free input, so a detailed explanation doesn't have to become a series of tiny recordings. Fast, accurate, dependable input serves one purpose: keeping your attention on the work instead of the mechanics of entering text.
How can you explain a long request and review it easily?
Speak your goal, context and feedback with Flow, then review the organized text. This example about AI products shows how continuous speech can be organized into separate paragraphs and numbered points.
A clearly organized message
Have you seen the launches of some of the new AI models and products over the past few days?
A few recent developments worth following:
1. The open-source Kimi K3 model has been released on Hugging Face, along with a technical report.
2. GPT 5.6 is also very strong.
3. Codex and ChatGPT have been combined. It's an OpenAI product, and it seems pretty impressive.
4. ChatGPT recently integrated Apple Health, so you can access your health data while chatting. That seems like a very useful feature.
Haven't a lot of AI health products come out in the first half of this year? Data is really at the heart of health products: once you have user data and context, the next question is how to use an agent to analyze it. If ChatGPT handles this, it can combine the health data recorded in Apple Health with your other context. That could be very valuable.
Products and models seem to be advancing very quickly lately. Also, the official release of DeepSeek V4 may be coming soon. Let's look forward to it.
Read the unstructured version
Um, have you seen some of the new models, AI models and product launches these past few days? Have they launched? Whether or not it was released, the open-source model is already on, released, and there's also a technical report you can look at. Right, and GPT5.6 is also very strong. Then codex and ChatGPT were combined, it's a product from OpenAI, and it seems pretty impressive. It has, right, recently ChatGPT also connected apple health, so when chatting with ChatGPT you can look at your own health data, which seems like a pretty useful feature. Weren't there lots of AI health products in the first half of this year? ChatGPT, actually, the core of health products is data, user data, context. Once you have that, what's left is how to use an Agent to analyze the data. So if ChatGPT does this, then it can take the health data recorded in Apple health and combine it with the other contact, and that could actually be very valuable. Right, and lately these product models, they're advancing very quickly. And an official version of something like DeepSeek might be coming soon too, so let's look forward to it.
Structure makes your own review easier
The output separates the product updates from the speaker's thoughts about health data. You can scan the names first, then review the reasoning, without hunting through one dense paragraph for every detail.
Accurate recognition reduces corrections; organized text reduces the work of rereading and editing. Before handing ideas like these to Codex or ChatGPT, you can more easily see what you have said and what you still want to add. Then ask questions, discuss the idea and move the work forward.
Can you look things up while speaking, or use your phone?
Explaining a task often means checking a document, looking at a page or reviewing the previous result. Flow doesn't require you to stay in the conversation while recording. You can look things up as you speak.
01 · Desktop
On Mac, press Fn and speak
With Flow's default shortcuts in the Codex desktop app on Mac, hold Fn to dictate and release it to finish. For a longer explanation, use Fn + Space to start hands-free recording, leaving both hands free to browse documents and switch windows as you add context.
Switching to another app doesn't interrupt recording just because you've left the Codex composer. On Windows, Flow shortcuts and hands-free mode likewise let you research and speak at the same time.
02 · Mobile
Walk and talk, without keeping the keyboard open
When using Codex remotely through ChatGPT on your iPhone, use the Flow keyboard to speak requests and feedback. Once recording starts, you can leave the conversation, dismiss the keyboard, browse other content, switch apps or even lock your phone and keep recording. Flow supports continued recording with the screen locked on both iPhone and Android.
Capture a follow-up while walking, or check another app while you explain. Moving to your phone doesn't mean reducing a detailed thought to a few short messages or typing everything on a small keyboard.
Let input follow your thinking, instead of keeping your attention on an input box. When you're done, return to the intended conversation, review and add to the organized text, then send it.
Why use Flow with Codex?
The question is about the input you repeat throughout the day: can you express yourself quickly, have your vocabulary recognized accurately and rely on a familiar workflow? Flow brings together fast voice input, a personal dictionary and text organization so you can spend less effort entering and checking what you want to tell AI.
Spend less effort entering text. More on the idea.
Add your product names and specialist terms to your dictionary. Then speak your next Codex request with Flow.