Coding agents can work for minutes while you move to another task. The problem is no longer watching them type. It is knowing when they finished and understanding the result without reopening every terminal.
Agent Callout 1.0 starts with a narrow answer: let the completed response call out from the Mac menu bar.
Recent agent sessions in one menu
Agent Callout reads the Claude Code and Codex session files already stored on the Mac. Recent sessions are grouped by project and marked as working or ready, making the menu a small status surface instead of another full dashboard.
Choose a finished session and Agent Callout extracts the latest final response, summarizes it in the selected Style, and speaks it in the selected voice.
Built for listening, not transcript playback
The launch release includes Casual, Professional, and Complete Styles. Casual and Professional compress the response for listening, while Complete reads the final answer without summarization.
Seven voices, adjustable speed, a global hotkey, and menu controls make the audio usable as a notification layer rather than a novelty demo.
A local-first boundary
Session discovery and transcript reading happen on the Mac. Only the final response selected for playback is sent to the authenticated service for summarization and speech, and transcripts are not stored by Agent Callout.
Usage is metered by audio time, matching the resource the product actually delivers. Signed releases can update in place without replacing a person’s local preferences.
Product decision
The principle behind 1.0
Agent Callout is not trying to become another place to run an AI agent. It owns the handoff after the work: notice the finish, hear the result, and decide whether the session needs attention.
That narrow boundary is how we choose what belongs in the product. A feature should make the handoff faster, clearer, or more trustworthy.