A few weeks ago you could record a voice message to your agent. Useful — especially for messy end-of-day debriefs and meeting recaps you'd never have the patience to type. But it was one-way. You talked, your agent listened, and whatever came back came back as text.
Now it talks back.
Open any chat, tap the phone icon, and you're in a real conversation. Your agent picks up, you start talking, and it replies out loud in a voice you picked in settings. You can interrupt it mid-sentence. And while the call is still live, it can send emails, check your calendar, look things up on the web, and call any of your connected apps — without breaking the flow.
It's the closest thing to actually ringing up a very capable friend. One who happens to have access to your inbox.
A quick call instead of a long chat
Sometimes I just want to know one thing before I get on with my day. "What's on my calendar today? Anything I should prepare for?" Typing that, waiting for the response, scanning it — that's a minute. Calling is ten seconds.
"Hey, what's my day look like?"
"You've got the product sync at 10, lunch with Priya at 12:30, and the design review at 3. The sync has a doc attached you haven't opened yet — want me to summarize it?"
"Yeah, two sentences."
Done, while I'm still pouring coffee.
Thinking out loud, with interruptions
I already use voice for messy thinking — talking through a problem is often easier than writing about it. But with pre-recorded voice messages, I'd talk for two minutes and get a response back. Good, but my agent couldn't interject.
Now it can. I'll start explaining something and it'll ask a clarifying question halfway through. "Wait — by 'the team' do you mean engineering or the whole company?" That small back-and-forth changes everything. It turns into an actual thinking partner instead of a transcription tool.
The best part is I can ask it to do something the moment I think of it. "Oh, and while we're talking — check if Maria is free Tuesday at 2 and send her a calendar hold." It reaches into my calendar, does it, tells me it's done, and we keep going.
While driving, walking, cooking
This is where it really shines. Voice-only contexts where typing isn't an option — commuting, walking the dog, making dinner while kids run around the kitchen.
- "Add 'more pasta' to the grocery list."
- "What did Tom send about the contract last week?"
- "Set a reminder for the dentist tomorrow at 3."
Each of these would be a 30-second detour if I had to unlock my phone and navigate to an app. On a call, they're part of the conversation — spoken in passing, handled in passing.
Pick your voice
Your agent has always had a personality in settings. It now also has a voice. Go to Settings → Agent and pick from a handful of options. Each one has a preview — tap play and hear your agent say a line in that voice before you commit. The voice persists across sessions, so every time you tap the phone icon, it sounds the same.
A small thing, but it matters. A consistent voice is part of the relationship.
How it works
Tap the phone icon in the chat input. Your browser asks for mic permission, and a second later your agent is on the line. Speak when you want, it replies when you pause. Tap End to hang up.
When the call ends, a short summary of what was discussed lands in your chat thread, along with any actions that were taken. Your agent keeps the full picture — nothing gets lost when you switch back to typing later.
Why it feels different
Most "voice assistants" are voice interfaces bolted onto a text system. You say something, they transcribe it, they process the text, they generate a text response, they read it out. There's a noticeable pause, a rigid turn-taking rhythm, and no tool calls while the line is open.
This is different. Audio goes in and audio comes out — your agent hears you directly, which means tone, pacing, and emphasis come through. Tool calls happen mid-sentence. Barge-in works because the audio path is bidirectional, not alternating.
It feels less like pressing a button and more like picking up the phone.