Why I Built a Notch App for My AI Coding Agents

Five agents. Twelve tabs. One is waiting. Which one?

I run Claude Code, Codex and Gemini CLI at the same time. Each one sits in its own terminal tab, and every few minutes one of them stops to ask for permission or a question. Finding which one meant Cmd+Tab roulette. So I started writing Notch Crew. It is a native macOS app that gathers every agent session into the MacBook notch. I wrote it for my own workflow first.

One line per session

Every session gets one row: status, context, tokens, model, and the mode for Claude Code, Codex and OpenCode. When an agent leaves shell commands running after its turn, its row says "2 shell tasks" instead of Idle, and it isn't marked done until they end.

Answer it where you are

When an agent asks for permission, the notch shows the same choices as the terminal, not a simplified Allow/Deny. When you press Install, it adds one small hook to each agent's settings and shows you exactly what it adds. Remove takes it out again.

You can also type into a session or drop a file on it. That works in Terminal, iTerm2, cmux and tmux. For any other app it copies your message and brings the app forward.

Hand it off

You can hand a conversation from one agent to another, and resume past sessions from Recent sessions.

Sessions with no window

Sessions the notch opens don't need a terminal window. I measured it: one real Claude Code turn, replayed into each host so every host got the same load. CPU and power come from the kernel's per-process counters (proc_pid_rusage, the same source as Activity Monitor's Energy tab), on a MacBook Pro M1 Pro.

Host	Extra power while an agent streams
Terminal window (Terminal, iTerm2, cmux)	26-77 mW
Session the notch opened, no window	0.1 mW

Terminal windows took 26-77 mW whether they were in front or covered. The notch itself, folded, uses under 1% CPU and 2-3 mW.

Stays on your Mac

It reads your agents' local session files to show their status. The only calls out are the update check and the purchase check.

Try it

Works with Claude Code, Codex CLI, Gemini CLI and OpenCode, plus the models Ollama has loaded. macOS 14 or later, Apple silicon and Intel. https://notchcrew.com

I'd love your feedback, especially from folks running multiple agents in parallel. How do you keep track of yours?