ICYMI: You can use Voice transcription in both Codex App as well as the CLI! 🎙️ Press the mic button or hit `Ctrl + M` and talk away! Available to 100% of the codex users :)

Can you share the gpt-6-7 / gpt-6 analysis? Thank you

You can use the same in the CLI by: 1. Enabling `voice_transcription = true` under [features] in ~/.codex/config.toml. 2. Focus the composer and press-and-hold Space to talk; release Space to stop and insert the transcription. Enjoy!

@OpenAIDevs What about Codex CLI Linux users ?

@meet__pandya @OpenAIDevs Also available in CLI (currently experimental):

@OpenAIDevs Tried on both windows and linux. Not working as of now, will wait for stable release

Now let me code using multimodal voice not speech to text but speech to speech.

soon :)

🎙️

It works really well! Hoping you can enable it in Linux soon. I love that you can also add to a prompt as well. You don't have to craft the whole message with voice, or even all at once. Excited to see how realtime convos work 👀

Why This Is Bigger Than It Sounds?

voice commands >>>

🥺🥺🥺🙏🏻

Comparison:

available for 100% of codex users, i see what u did there xD

Nice! Gotta test it out vs using Superwhisper!

is there a strong reason not to standardise on the space bar across cli and app for this? a little bit less friction goes a long way

Valid ask, let me see what was the reason for it! cc: @ajambrosino @edbayes for vis

please add /compact to the app too!!!! 😭😭😭😭

@reach_vb Consider this approach. In my Web CLI I show realtime transcription and a smaller waveform so we can see if it is getting it correct. It also helps if I want to type paths or insert images mid conversation.

the underrated use case isn't dictating code, it's narrating intent. voice is terrible at symbols but great at context — 'handle nulls from the api, don't break the retry loop' lands better spoken than typed when you're talking to an agent

That is a fantastic update for the workflow. Combining intellectual curiosity with tools that reduce friction is exactly how we make things better....

I still prefer Wispr Flow because it remembers all my weird pronunciation with my accent.

@OpenAIDevs What about the codex ide plugin?

@OpenAIDevs In on windows PC i get the goice activation in the CLI but it doesnt record anything.. dont know what to do..

I use Handy for that

@OpenAIDevs Thank

We want remote development via the codex app

@OpenAIDevs when ide？

Windows???

sooner than you'd think

voice input in a coding CLI is one of those things that sounds gimmicky until you actually use it. describing what you want in natural language while staring at the code is way faster than context-switching to type out a prompt. especially for those "refactor this function to handle X edge case" type instructions where talking is just more natural than typing

can you invite me to the "gpt-67-analysis" please?

It's not working no CLI running on Windows.

Voice is the future If you want to vibe code like you're in 2036 try out Always on, local transcription - feels like magic It's blazing fast as well, if you couldn't tell by the name Don't just take my word for it tho:

Cli ?????

You guys still have not fixed the 403 Forbidden error. Voice Transcription is still not working because of that. And i think Eric closed all github issues related to that because ""...under development" are not yet ready for use."

Does it count towards usage limits?

Love this rollout—voice-to-code is especially useful for drafting prompts while context switching. A small tip: punctuation voice commands make CLI transcripts much cleaner for follow-up edits.