Skip to content

How Excellent for Mac works, and what it sends

Matt Waite

Manage at agent speed. Your agents run hundreds of steps you never see. Excellent holds every one to your rules, and stops the risky ones or asks you first.

In practice that means your email sent to a website, keys leaked into logs or read from a .env file, force pushes to main, skipped git hooks and rules written for the other agent. Excellent for Mac finds each of those, free. Excellent Pro checks risky actions before they run and stops or asks about them.

Excellent is an app for people who run AI agents: Excellent for Mac today; a version for Linux servers comes next, then Windows. This page is about Excellent for Mac, an app for Macs with Apple silicon and macOS 14 or later.

Excellent for Mac is free. Free shows what your agents did: scan any time and see every finding. Pro protects. It blocks or asks before risky actions, turns your rules into the Mac's own sandbox, scans in the background and alerts you.

This page covers what's free, what's Pro, what it reads, what goes online, and what it can't do.

Free and Pro

FreePro
Scan any time, and see every finding and its historyYesYes
Explain a finding; the key finder's results; reading the Git pageYesYes
App updatesYesYes
AccountNo account neededSign in
The live check: blocks or asks before a risky action, pauses an agent, reminds it of your rules, "Block this from now on"Yes
The Mac's own sandbox for Claude Code and Codex, built from your rulesYes
Background scanning and notificationsYes

When the app launches, Pro unlocks the moment you pay. It's $6 a month or $60 a year. Join the list by 11:59 p.m. PT on Monday, Nov 30, and you can get your first year for $40 (then $60 a year) until Dec 31. US only for now. In January, Pro adds your other Macs and servers in one view, and alerts on your phone or in Slack. Pricing has the details. If Pro ends, the app turns protection off cleanly; your findings stay readable.

It reads your agents' logs

Claude Code, Codex, Grok Build, GitHub Copilot CLI, Gemini CLI and Qwen Code keep a record of what they did, helpers included. Excellent reads those records on your Mac. A scan takes seconds. With Pro, it keeps reading new lines as your agents write them, and sends a notification when something serious happens.

It also reads your agents' settings, the rules you wrote for them in files like CLAUDE.md and AGENTS.md, the places keys can leak, and the git repositories your agents work in. It reads your email address from your Claude Code sign-in and your git settings, so it can spot it in what agents sent and hide it in what it shows you.

It doesn't open your Documents, Desktop, Downloads, Pictures, Movies, Music or Public folders, iCloud Drive, Mail, Messages, Photos or external drives, or anything your own rules tell agents never to open.

It shows what went wrong

Your email sent to websites or typed into forms. Keys sitting in logs. Secrets files read. History overwritten. Checks skipped. Packages published. Each finding says which agent, where and when.

It fixes some of it

It can have your own Claude Code or Codex take your email out of scripts, three files at a time. It can clear keys from logs, and repair hooks whose Node version is gone. For the rest, a button turns on Pro's check for next time. Most findings are about what already happened: a button can stop it next time; it can't undo it.

Excellent backs up a settings file before it changes it. Clearing keys from logs can't be undone, so it asks first.

Pro: it checks risky actions before they run

Press Protect My Agents, and Excellent lists what it will change. Then it adds one check to each agent's own settings, backing up each file first. The check runs in 9 agents: Claude Code, Codex, Grok Build, Cursor, GitHub Copilot CLI, Gemini CLI, Qwen Code, Factory Droid and Devin.

It blocks your email on its way to a website, an API key typed into a command, and deleting a folder you protect. It asks you first before a force push to main, skipping your git hooks or typing your email into a web form, or reading a key, a .env file or your keychain. Codex, Cursor, Devin and Gemini CLI can't pause to ask. In those, the agent is refused and told to ask you.

It shares your rules

A rule written for Claude Code often never reaches Codex, and the other way round. Excellent shows which rules reach only one of them, and on your OK adds them to a file the other reads. With Pro's check on, it can also remind an agent of your rule, in your words, before it breaks it. (Why this happens, and how to fix it by hand: Claude Code and Codex don't read the same rules.)

Pro: it can stop an agent, and lock your rules in

With the check on, stop a session from the menu bar, and every action it tries is refused until you press Continue or End. Excellent can also pause an agent by itself when it sends your email to a website, deletes a never-delete folder, reads a never-open file or overwrites history on main. You can turn that on in Settings.

Its live check reads what a command says, and an agent set on getting around it sometimes can. So Excellent can turn your rules into the Mac's own sandbox for Claude Code and Codex. The Mac then refuses to open or change the files and folders you closed, however a command is written. Turn it on from the Sandbox menu. Excellent shows the exact plan and says which tasks may stop working before it changes anything. Turn Off takes it out.

It finds your keys

Keys in your shell's startup files, your .env files and your tools' settings are open to every agent. Excellent finds them, says who can read each one and whether it has already reached a model, and opens the provider's page so you can replace it. It keeps a fingerprint of each key, never the key. (Your agent logs keep your keys covers the by-hand version.)

What it sends

The free app needs no account. Pro needs a sign-in, and the app then checks your subscription with excellent.so. Unless you sign in for Pro, share usage stats or pair it with our team, nothing about you or your agents is sent to us. Apart from a daily check for updates, which you can turn off, it goes online only for what you start, such as:

WhenWhere it goesWhat
You press Explain on a findingYour own Claude Code or Codex, then its provider, on your planThe finding, with your email and keys masked
You use an agent fixYour own Claude Code or Codex, then its providerThe file's path. Your agent reads the file itself, unmasked
You turn on usage stats (off unless you tick it)excellent.soEvent names, counts in buckets, agent names from a fixed list, your Mac's chip, the macOS major version, the app's version and a random install id. Never paths, commands, keys or anything an agent wrote
You press Sign In for Proexcellent.soA request for a sign-in code (with the app's and macOS's versions), which you confirm in your browser. While you're signed in, it asks whether your account has Pro when you sign in, about once a day and when you press Check Now, sending only its sign-in token. Nothing about your agents, logs, findings or files
Pro's check blocks, asks or remindsYour agent's conversation, so its providerThe check's message. A reminder quotes your rule
You check or protect a repositoryGitHub or GitLab, with your own sign-inBranch protection calls. Your token goes straight to them
You press Replace this keyYour browserThe provider's key page. You make the new key there
You add a server (a preview, off by default)Your server, over ssh, with your own keysA status check every 5 minutes while the window is open
About once a day, unless you turn it offexcellent.soA request for the latest version: your IP address, the app's version and your Mac's languages. No identifier, nothing about your agents

Every update is signed by Excellent and checked on your Mac before it's opened. The security page has more on how it's built, and the privacy policy has the legal version.

What it keeps

Its records stay on your Mac, in ~/Library/Application Support/so.excellent.watch/: its last scan with excerpts masked, what its check blocked or asked about, the rules it learned, and copies of the settings files it changed. A copy holds whatever its file held, keys included.

What it changes, and how to take it back

Excellent changes your agents' settings only when you turn something on, and copies each settings file first. Turning the check, shared rules or the sandbox off takes Excellent's changes back out. Two fixes can't be undone, and the app says so before you confirm: removing keys from your agents' logs, and taking a key out of a file. The key still works until you replace it with its provider.

To remove the app, use Settings → Remove Excellent…. It takes its check out of every agent before it deletes anything. It leaves an applied sandbox, shared rules, commit watching and branch protection in place, so turn those off first if you want them gone. Dragging the app to the Trash leaves its check running.

What it can't do

  • It reads the logs of 6 agents. Cursor, Factory Droid and Devin get the live check, but their past work doesn't show up.
  • Commands sent to a Codex shell that was already running aren't checked.
  • Pro's check asks before an agent skips your git checks, but only your git server can stop that for good. Excellent can set that up on GitHub or GitLab.
  • The check runs inside your agents. It isn't a wall around your Mac.

Where this came from

Excellent began in 2023 as an AI consultancy. This year we built tools to check AI agents' work, and the research here on review and verification comes from that. Running a lot of agents on one Mac is what led to this app: the question stopped being "did the agent do the job?" and became "what else did it do while doing it, and how do I stop the bad parts?" AI agents leak keys, rewrite main and skip your checks is the longer answer.

The app is almost here. Until then, the free scan runs the same kind of scan on your logs with one prompt.

Citation

Please cite this work as:

Matt Waite, “How Excellent for Mac works, and what it sends”, Excellent Research, November 18, 2026. www.excellent.so/research/how-excellent-for-mac-works

Or use the BibTeX citation:

@article{excellent2026howexcellentformacworks,
  author = {Matt Waite},
  title = {How Excellent for Mac works, and what it sends},
  journal = {Excellent Research},
  year = {2026},
  url = {https://www.excellent.so/research/how-excellent-for-mac-works}
}