AI coding agents are most useful when they can keep working while you move on. That only works if you know when they finish. Version 1.6 expands Agent Callout to Cursor and makes the finished-response loop more automatic for new installs.
The goal is simple: one quiet place to hear what your agents completed, no matter which supported tool did the work.
One menu for three agent environments
Agent Callout now reads recent sessions from the Cursor IDE agent and the Cursor CLI, then places them in the same menu as Claude Code and Codex. Cursor support does not require an editor extension or a separate install inside Cursor.
The details still matter. Sessions stay grouped by project, finished work is separated from work in progress, and only the response being voiced leaves the Mac for summarization and speech.
Finished work can speak for itself
Auto-play is now on by default for new installs. When an agent finishes, its response joins the playback queue. If another callout is already speaking, the next one waits instead of talking over it.
The off switch remains obvious in Settings. Turning auto-play off returns Agent Callout to click and hotkey playback without removing either control.
A broader agent layer, not another inbox
Adding another agent could have meant adding another panel, notification stream, or workflow to manage. We chose the opposite. Cursor enters the same small menu-bar surface and follows the same playback rules as every other supported agent.
That keeps the product focused on the handoff between an AI agent finishing and a person understanding what happened.
Product decision
The principle behind 1.6
Agent support should follow the developer, not force the developer into one tool. The local session adapter is the product boundary: once an agent can be understood there, it can share the same menu, privacy rules, and voice experience.
That is how we will evaluate more agent integrations: does the new support reduce context switching without creating a new place to babysit?