Talk live

A spoken conversation with The Machine about a to-do, with OpenAI's voice. It works in the to-do's chat, and sends a draft only after you say yes to it.

1. Start a conversation
2. Talk through all your to-dos
3. Ask for things, and send with a yes
4. Pick a voice
5. A record in the chat
6. What you need
7. When it doesn't work

Start a conversation

In a to-do's chat, tap the wave next to the mic. Allow the microphone if your browser asks. When the bar says Live, talk as you would on a call. The Machine answers out loud, and you can talk over it to interrupt.

It knows the to-do you're in: its messages, its draft and its chat. It can also look up your other to-dos.

The bar has Mute and End. Say goodbye, or that you're done, and it ends by itself. A conversation lasts up to 10 minutes. A minute before the end, the timer shows how much is left. Start it again to go on.

Talk through all your to-dos

Tap Talk in the sidebar (in the menu on a phone) to go through everything that needs you, by voice. The Machine says how many are waiting and takes them one by one: Ready first, then To decide. It sees your other to-dos too, like the ones still Preparing, and tells you when one is ready.

Talk has its own page: one conversation from start to end, with the to-do it's talking about on top, its draft, and Open the chat. It doesn't move you from chat to chat.

Ask it to read a draft, change it, snooze it, mark it done, or move on to the next one. Sending works as in a to-do's chat: one draft per question, after a clear yes. What you say about a to-do goes into that to-do's chat. When a new to-do comes in, it mentions it at a pause. Say "add a to-do to..." and it adds a new one, without leaving the one you're on.

It reaches the rest of The Machine too. Ask about a person, an organization, a project, a meeting or a promise, and it looks it up and shows it. It can snooze, mark done, dismiss, rename or reopen a to-do, also one in Done / dismissed. For anything else, it adds a to-do, lets its chat handle it, and tells you the answer.

Meanwhile, other pages are fine: the bar stays at the bottom, and Talk in the sidebar brings you back. This conversation lasts up to 30 minutes. It keeps going in the background on computers and Android. On iPhone, it stops when the app leaves the screen.

Ask for things, and send with a yes

When you ask for something, like "write a reply saying Thursday works" or "snooze this until Monday", The Machine hands it to the to-do's chat, as if you had typed it. The chat follows its usual rules: it can change the draft or snooze the to-do.

The chat can also look things up, like a store's opening hours on the web. It shows Working on it… meanwhile, and you can keep talking. When the answer is ready, The Machine tells you, and where it found it. Ask something else meanwhile and it waits its turn.

To send a draft, say so. The Machine asks first, naming who it goes to and how, like "Send this reply to Lotte by email?". It first reads the whole draft out: who it goes to, the subject and the text. So it works with your phone in your pocket, too. Answer with a clear yes, in any language, and that one draft goes.

• It sends exactly the draft it read out. If the draft changes after the question, it reads it again and asks again. A question you don't answer within 3 minutes expires.
• If your answer isn't a clear yes, nothing is sent and it asks again. "No" sends nothing.
• A yes always sends one draft. To send several, use the follow-up review.
• The chat shows Approved by voice with Undo for a few seconds before it goes.

Tapping Approve works too.

Pick a voice

In Settings, General, under Voice input, pick how The Machine sounds at Voice. Tap play to hear a sample. Cedar is the default. ChatGPT's Cove voice isn't available for Talk live.

A record in the chat

What you both say is written into the to-do's chat, under a Talked live line. You can read it back later.

What you need

Talk live uses OpenAI's realtime voice, so OpenAI has to be set up in OpenClaw:

• A ChatGPT sign-in: included in your plan, where your plan has access.
• An OpenAI API key: OpenAI charges your key per minute of talk, and the bar says so. When both are set up, OpenClaw uses the key.

Settings, General, Voice input shows whether Talk live is ready. Without OpenAI, the wave doesn't show. Your audio goes from your browser straight to OpenAI. Your key never reaches the browser.

When it doesn't work

• The microphone is blocked: allow the microphone for this site in your browser's settings.
• Talk live couldn’t start: check OpenAI in OpenClaw, then try again.
• Talk live lost its connection: start it again. What was said so far stays in the chat.