The context
I run several Claude Code and Codex sessions at once, across several projects. One is waiting for a permission, another finished ten minutes ago, a third is asking a question: I only find out by jumping from window to window.
The challenge
Knowing at a glance where each agent stands, and answering it without leaving what I’m doing. No piling notifications, no extra window.
My answer
The MacBook notch becomes a black island that extends it. Compact, it shows who’s working and what’s waiting. An alert drops from it when a session needs an answer; hovered, it unfolds with the command to allow or the question to settle. On hover, every session, sorted in the order to act.
The result
You allow a command, answer the AI’s question or clear what’s done, right from the notch: Claude Code gets the answer and carries on. A native app, designed down to the point, every spring and every fade, then built with Claude Code, in SwiftUI.
What Notchi can do
Show at a glance who’s working: Claude’s star, Codex’s cloud, or both
Drop an alert when a session needs an answer or is done, with a discreet sound
Allow or deny a command without opening Claude, from the alert or the list
Answer the AI’s questions in one click, without leaving what you’re doing
List every session in the order to act: waiting, done, running
Open the right session in Claude or Codex, or the link the answer gives
It all happens under the notch.
Notchi’s island, as it drops from a MacBook notch. Hover anything, click anything.Notchi’s island, as it drops from a MacBook notch. Tap it: everything works under your finger.
Move over the notch (or tap it): the list of sessions opens. Allow the command, answer the question, clear what’s done. The sessions keep working, and nothing touches your computer.
No more switching windows.
Four moves, all from the notch.
Who’s working, at a glance.
Compact, the island shows who’s at work, Claude’s star, Codex’s cloud, and how many sessions are waiting or done.
Allow without opening Claude.
A command to approve? Deny or Allow, from the alert or by hovering it in the list. Claude gets the answer and carries on.
Answer its questions.
When the AI hesitates, its options show up under the notch. One click and it’s back to work.
The summary, once it’s done.
What was done, in a few lines. Open the session in one click, or clear it.
