Voice, notifications, and notes
The assistant's speech speed now has its own control under Settings → Conversation, starting at a natural pace. Playing back a response uses the selected speed. The control that already existed under Voice stays in place and now states that it applies to alerts.
Longer responses are no longer cut off before they finish.
Listening confirmation
The spoken confirmation played when the microphone opens can be turned off or replaced with a word of your choice, under Settings → Conversation. It remains on by default and confirms that the microphone is live.
Commands limited to the current workspace
The close everything and close the agents commands now apply only to the workspace in use. The others stay open. To act on a panel in another workspace, name it.
Focus that follows across workspaces
Asking to focus an agent located in another workspace now switches the view to that workspace. The same applies to the command that takes you to an agent.
One response per request
The assistant performs the action and reports only the result, without announcing what it is about to do. Responses now describe what happened instead of merely confirming execution.
Sound on alerts
Agents that ask a question, wait, or finish now play a short sound when the app's voice will not speak. Outside the app, the first alert in each batch also arrives with sound; the following ones update the summary silently.
Agents in another workspace now raise alerts the same way, and the action that takes you there opens the matching panel, switching workspaces automatically.
Per event notifications
Under Settings → Notifications, each event (finished, waiting on you, interrupted) has its own controls: the on screen card, its sound, the spoken readout, and the system alert with its own sound. A master control mutes every sound at once.
Notes per project
Every project now has notes, opened with ⌘K or by voice. Text is saved automatically as you type. You can keep several notes, switch between them from the menu at the top of the panel, name them (or let the first line become the name), and open several windows at once. Deleted notes can be restored from the trash in the same menu.
The assistant and the agents can write to notes on request. Deletion is reserved for the user, from the panel, and their access to notes can be turned off under Settings → Tools.
Voice command model
Under Preferences → Voice, the Command model selector defines which model interprets the spoken phrase, with accuracy and response time listed for each option. The current default answers in about one second in the heaviest cases, compared with more than six seconds in the previous version.
The command key now keeps capturing for half a second after it is released, preserving the last syllable. When a phrase is not understood, the app says so instead of guessing at the meaning.