Real-time voice chat
GroupGPT rooms aren't text-only. You can drop into a live voice call with everyone in the room and keep talking while the chat — and Cortez — carry on alongside you. The calls are peer-to-peer over WebRTC: audio flows directly between participants, and the server acts only as a signaling relay to introduce people to each other. That keeps latency low and means your conversation isn't routing audio through a middle box.
It turns a room into something closer to a workspace than a thread — hop on voice to talk through an idea, then fall back to text and the AI without ever leaving the room you're in.
Voice memos
When a call is more than you need, leave a voice memo. Record an audio note right in the composer and drop it into the conversation, where anyone can play it back inline. This round tightened the round-trip — more reliable capture, a smoother handoff, and fewer dropped recordings — so a quick spoken thought lands cleanly instead of getting lost.
Translation for every mode
GroupGPT's per-user translator mode lets each person read the conversation in their own language — messages are fanned out through the TranslationService and delivered translated, in real time, only to the users who asked for it. Until now that worked cleanly for ordinary messages but lagged behind the product's other chat modes.
This update wires translation through every mode. Whether the AI is answering normally or working in thought mode, translator users get the same live, language-matched copy. One person types in English, another reads it in Spanish, a third in Japanese — and nobody changes how they write.
A cleaner mode picker
The chat composer's modes — the different ways GroupGPT can respond — moved into a dedicated mode picker. Instead of competing for space in the toolbar, modes live in one place that's easy to scan and switch between. It's a small UI change that makes the product's range of behaviors discoverable rather than hidden.
Thought-mode pipeline fix
Thought mode lets the AI reason through a problem with its working visible, then deliver a final answer. A bug in how message types were threaded through the pipeline meant the intermediate and final pieces could be collected out of order. That's been corrected so thought-mode responses assemble in the right sequence every time.
Why it matters
Voice closes the gap between a group chat and a real conversation: jump on a call when you need to talk, leave a memo when you don't, and let the text thread and the AI keep working the whole time. Pair that with translation that now holds across every mode, and a mixed-language team can talk, type, and read along together — no separate call app, no separate translator, no leaving the room.