Alerts with a natural voice

Spoken alerts now use the same natural voice as Command mode. You can pick from eleven voices and adjust the volume and speed of each. The volume applies to the voice only, so lowering it does not affect audio already playing. A button next to each voice, both for alerts and for the assistant, plays a sample phrase with the current settings.

Speaking over an alert stops it immediately. Phrases such as “enough” or “stop talking” end it without waking the assistant. If what you said contains an actual request, the conversation opens already aware of the subject.

Cancelling with Esc

The Esc key cancels any speech in progress: a dictation started by mistake, an abandoned command, a conversation opened at the wrong moment, or a spoken alert. Text being dictated is discarded and nothing is sent to the agent. The key works even with CanvasCode in the background, and is captured only while speech is in progress.

Work areas

A project can be split into areas, shown as tabs, each with its own panels, layout and camera position. Switching areas ends nothing: agents keep working, alerts keep arriving, and the conversation and terminal stay exactly as you left them.

The + button creates an area and asks for its name, clicking the current tab renames it, and ⌃⇥ moves to the next one. Search with ⌘K and voice commands reach panels in every area, including those off screen. Existing projects get an area named Main containing everything.

Wallpaper

Besides images, the project background accepts GIF and video (MP4, MOV, M4V), looping, muted, pausing automatically when the window is not visible. A control sets how much the background is darkened, with a preview alongside.

When you ask the assistant for a wallpaper, it tells you once the image is ready and asks whether you want to keep it. Requesting another option is immediate, because the remaining results from that search are kept.

Dictation

Voice keys now apply only to the window in the foreground, and no longer fire with the app behind another window. While dictating, the indicator shows the name of the agent that will receive the text and the last part of what you are saying. With no agent selected, the text is kept and the screen asks who should receive it.

If the window loses focus, the microphone stays active. After a period of silence, dictation pauses and you choose between continuing, sending or discarding, and resuming keeps everything already said.

Dictation history

The last twenty dictations appear in a list in the footer, with the text, the time and the recipient. You can send the same text to another agent, or copy it, without repeating yourself. Resending also works by voice, identifying the dictation by the subject you mention.