正在加载视频...
视频加载失败
Voice is now available in your existing Codex threads. Talk through a PR with the agent that wrote it, debate the architecture, or work out what comes next. You can jump into voice whenever something needs a back-and-forth, then let the agent keep working once you’ve figured it out together.
37 条评论

I recommend switching to light reasoning for the live back-and-forth. Talk through the plan, then turn reasoning back up before sending it off to work.

Oh and new voice chats also now respect your project and model selection.

Omg yesssss The one voice chat orchestrator pattern was getting odd for some things

What’s the benefit of this over one master voice chat that delegates to other threads and orchestrates them? This lowkey seems convoluted and complex outside of a few edge cases. My main chat workflow has been legit though.

Both are very useful workflows imo. I prefer to use an orchestrator voice chat to plan and dispatch new threads. Then, if I happen to have a lot of task-specific context to follow up with, I find it easier to talk directly to the worker agent as oppose to playing telephone through the orchestrator.

True, I was asking for this back when voice first came out, but then I found the workflow above. Also, would it be possible to have voice open in Codex and something like ableton in the background? Apple makes it so they switch to mono w/mic or something and cuts stereo audio 😳

This is great!

It seems like there’s no way to use Voice with regular Chat limits on PC anymore, which feels like a step backward.

so this is different from the previously small model that ran the voice session before?

that’s so much needed THANK YOU

Great news, thanks! Here is a suggestion: add voice persona to each project. Each project has a pm voice persona, for example Peter, Paul and Mary. Each have their own distinct voice and style

this is great @guinnesschen! can we also get read-aloud in Codex?

I ran into issues while using voice when driving, because the connection broke and I couldn't resume the chat (voice icon disappeared). I was using the ios app, remote connection to home codex.

Codex voice would be really awesome if the processes behind it wouldn't be that slow. Voice itself is really great. Congrats for that. But i fall asleep getting this simplest things done.

I'm a Codex Ambassador hosting (what I think is the first) event that's fully focused on voice - would be sick to have you demo your workflows! I only have myself and one other speaker so far (waiting to promote it widely until I line up speakers)

Needed this. Voice in the actual thread seemed more effective to me. And losing voice once I ended the initial voice session was frustrating.

Not on remote project? kind of big deal no?

@rudrank This is a big unlock!

On mobile there’s a bug where if I invoke voice mode mid conversation the voice chat responds likes it’s a brand new conversation, it’s very frustrating

voice in codex threads is actually huge

what happens if I start voice while the agent is mid-turn? does it interrupt the run, queue the transcript for the next model boundary, or open a parallel conversation with the same context? that determines whether voice is only for planning or can steer a live worker.

oh wow - this is exactly what I was looking for! Thank you! I hope I can avoid messy steering threads now.

so rad

With Astra 🚀🚀🚀🚀

nice change, i slightly feel like the chats should be separated so it doesn't bloat the current chat

I was literally looking for this feature yesterday, well done! And it's also on mobile too!

This is the right use for voice. I don’t want to dictate code, I want to argue with the agent before it writes it.

huge

@DarwinSantosNYC Really excited for this

@ajambrosino Thank you, this is what I’ve been asking for

I've been waiting on this! Thank you @guinnesschen ! Makes this tool even more magical.

I used this yesterday not knowing it was out and honestly, one of the best experiences I’ve had with AI so far. Makes me excited for the future!

.

amazing work

awesome. OpenAI really is on fire here. the pace at which you listen to feedback and adapt your product accordingly is just amazing :-)

YESSSS

Nice work cutting out the middle man. It would be great if you supported a mode where it would not announce every action it is going to take. This was all to hear the date.

