I wanted to highlight how powerful ChatGPT Voice with Codex is. This is how I operate my computer now. I just ask for things, and things happen. There is no limit to your use cases! The future is already here!

Voice sucks I hate it It needs to be open and local and part of the codex app such that I can switch the infrastructure between AI providers

Build it

Actually pretty sure you can do something with the tools in the codex app and another voice provider maybe?

I already have it set up with some tmux infra monstrosity I SSH to over my phone

Oh yeah also this video is subbed with Codex by just asking Codex and selecting it in Finder and asking the voice to take an AppShot

Why is codex voice personality way more monotone then chat live voice mode?

It's great for control, but weak in collaboration. I find myself switching to voice transcription via Handy in continuous listening mode instead of using Codex Voice simply because I don't want that bridge between the voice model and the strong worker model. There's massive delay and forwarding issues of complex content because communication is not seamless. I do think the current mode is already a milestone and I like it, but it doesn't yet solve my main problem: voice-to-voice collaboration with agents.

But if I start voice from my project like so, then its working directory is not my project's directory, its some realtime-voice-chat-x temporary directory and it has no clue about the project. Is that intentional ?

Yes. The Voice is an orchestrator, it can access anything, any projects, any folders etc... just ask

hmm but when you start voice from the project (my first picture), then its working directory should be the project's! It's currently very confusing. If it's not intended to be that way then don't put the voice icon there.

I understand the orchestrator thing and I like it. But if you start from a project there it should imply that we're going to orchestrate things in that project by default.

it'll still be launched the same regardless right now

Voice is phenomenal!! The only thing I don’t understand is why a new conversation/thread must be started every time just to use it, and it is not possible to forse it up in an ongoing thread instead

Love it. How can I get the avatar out of codex and on the screen?

The orb should automatically stick around wherever you are if the voice thread is not selected

My struggle with voice is that I don’t have a constant stream of stuff to talk about. I can kick off a bunch of work, but that takes time to complete. Then work needs to be reviewed. I want to listen to music and talk out loud without a live mic. There’s no quick mute shortcut.

How do I know what model it's using? I tried it and couldn't find the model it was using at the output was terrible. All I wanted was for it to build a PowerPoint. Regular computer use worked better. No idea why. I assumed because it was using a lower model.

i love using voice to text, and you’re ahead of anthropic by quite a bit on transcription accuracy but i can’t bring myself to use voice. doesn’t feel natural

Checking..

@victornunez What about usage limits? Does it take a lot, or is it similar to normal GPT 5.6 sol usage, for example?

@dkundel We have come so far 😊

how is your use rate though? My allowance used to last forever. seems like after this voice release update, i cant even get a few days out of my sub. It just horrible. I even did the luna worker model trying to reduce burn. still used 25% in 2 hours of use today. I'm Sad

Loving codex + ChatGPT lately, very well done to the team. Moving a lot of my personal agentic setup to codex

this is definitely running on cerebras, it’s not as fast on mine

It seemed sad when you said now delete them all.

I know even the voice felt it :(

Los tokens son los límites.

Wow! Just imagining what’s possible w/ next gen or 2 models, where they’ll be good enough to join meetings etc as (mostly) passive observers and take actions on your behalf. Super excited for the “multiplayer mode” era…

The magic is real. The leverage comes from knowing when the agent should ask, act, or stop.

When you start working with Voice in Codex, it just blows your mind. I love it! But there is some feedback from my side: I am struggling to let voice kick off new Threads under the right project. When I start a new Thread via the Sidebar on a project, it always creates a Voice...

Voice plus agents could be the moment AI moves from a tool on your screen to a true assistant.

this is the kind of concrete tip that sticks

i use "Shift + E" to toggle the live voice mic. I might be using live voice for one session, but also using speech to text in another session. The only way to know if mic is on/off on live is by hovering over it. would be great if it showed a "mic off" indicator

Please bring Standard Voice to Codex/Work modes.

curious how far we can push it. adding it to the list

Last night i was in bed and wanted to see what it was like not having to get back up to go to my desktop. No more stressing my neck and back more than I have to.

The floating Codex Voice overlay is super laggy btw. Like, just moving it around is super super slow. Macbook Pro, M5 Pro, 48GB.

Thats actually crazy , what do you mean I can run live server , control databases , clean space in my Mac , all without touching the keyboard

bro codex created it in parallel while i was watching 🤣 forget to mute my mic

That sounds great, I learned something new here. I'll use ChatGPT voice to communicate with Codex from now on too.

👀🤔

bro .. so many use cases and you do a create and delete folder ok

New issue: ran out of battery on my phone while asking for things. Also. Trading in my MacBook Air for more ram - max mini. Here I come.