GUIDE

Codex notifications on your Mac

To know the moment a Codex session finishes or stalls, you need something watching outside the terminal - because a Codex run buried in a background tab will happily sit done, or waiting, for an hour while you work on something else.

Why the terminal alone won’t tell you

Codex runs in a terminal tab, and a terminal tab is silent by design. If it’s behind another window or on a second monitor, a finished run and a stalled one look identical: nothing moving. You end up checking it every few minutes, which is the exact interruption an agent was supposed to remove.

What NapCat notices, without a plugin

NapCat recognizes Codex among 27+ coding agents by reading your Mac’s running processes - locally and read-only, nothing to install into Codex itself. From that it can tell the two states that matter:

  • Working - the Codex process (or its subprocesses) is burning CPU. The notch shows a blue Working… bar so you can see it’s alive.
  • Finished - activity drops off. NapCat plays a soft chime and surfaces the row, so a long run announces itself instead of waiting to be discovered.

The notch and the sounds

When a Codex session changes state, the Dynamic Island-style panel opens by itself and the session’s row moves to the top: what it was doing, which terminal it’s in, and how long it ran. A quiet sound marks the finish so you can stay heads-down in another app and still catch it. Click the row and NapCat brings that exact terminal tab forward in Terminal, iTerm2 or Warp - no hunting through windows for the run that pinged you.

An honest line on what’s Claude-specific

Two NapCat features are tied to Claude Code today, because they lean on Claude Code’s hooks: the Allow/Deny permission approvals in the notch, and the usage meter. Codex notifications here are the activity-based kind - finished and working, detected from the process, not a permission-prompt approval you can answer from the menu bar. We’d rather say so than imply Codex gets a button it doesn’t.

Everything Codex does get - live monitoring, jump-to-tab, notifications and the nap overlay - is on the Codex page. If you also run Claude Code, its notifications go a step further in Claude Code notifications, and long unattended runs are covered in run Claude Code overnight.

Questions

Does NapCat notify me the instant Codex finishes?

It fires within a few seconds. NapCat samples the process list on a short interval, so when a Codex run's activity drops off it chimes and opens the notch almost immediately - fast enough to switch back before you've lost the thread.

Do I have to configure anything inside Codex?

No. Recognition is by process, so a Codex session appears the moment it starts. There's no plugin, hook or config file to add to Codex - NapCat just reads that the process is running.

Can I approve a Codex permission prompt from the notch?

Not today. The Allow/Deny approvals are Claude Code specific because they use Claude Code's Notification hook. For Codex you get finish-and-working notifications and one-click jump to its terminal tab to answer prompts there.

Will Codex notifications wake my Mac or send data anywhere?

No. NapCat is local and read-only with no telemetry and no account. Notifications are just the menu-bar app reacting to processes on your own machine; nothing about your Codex work leaves the Mac.

NapCat does all of this from your Mac’s notch - for five dollars, once.

Download - free 2-day trial