Documentation

Getting Started

Requirements

• Mac with Apple Silicon (M1, M2, M3, M4 or later)
• macOS 13 (Ventura) or newer
• ~4 GB of available RAM
• ~2 GB of disk space

Installation

1. Download Sentinel.dmg from the link provided in your early access email.
2. Open the DMG and drag Sentinel to your Applications folder.
3. Open Sentinel from Applications or Spotlight.
4. On first launch, follow the setup wizard to grant Full Disk Access.
5. Sentinel downloads its AI model once (~700 MB). After that, it works fully offline.

Granting Full Disk Access

Sentinel needs Full Disk Access to read your Messages database. This is a macOS privacy requirement. Your messages are only read locally by the on-device AI model.

1. Open System Settings on your Mac.
2. Go to Privacy & Security > Full Disk Access.
3. Click the + button.
4. Find and select Sentinel from your Applications folder.
5. Toggle it ON and restart Sentinel.

How It Works

Sentinel runs a purpose-built AI model entirely on your Mac. When a new iMessage arrives:

1. Sentinel detects the new incoming message from your local Messages database.
2. The on-device AI analyzes the message using an 8-step behavioral rubric.
3. If manipulation or deception is detected, you get an instant notification with a breakdown of what was found.
4. All flagged messages appear in the Sentinel dashboard with full analysis details.

What Sentinel looks for

Sentinel's rubric walks through each message checking for:

• Identity misrepresentation - fake identity or situation to support an ask
• Manufactured urgency - artificial time pressure with exaggerated consequences
• Emotional coercion - guilt, shame, or obligation used to force compliance
• Threats - social, financial, or reputational harm to control behavior
• Phishing - requests for passwords, codes, or sensitive personal info
• Financial fraud - scam patterns in money or resource requests
• Masked threats - intimidation hidden behind "lol" or "jk"

What Sentinel ignores

Sentinel is specifically trained NOT to flag:

• Sarcasm and jokes between friends
• Genuine emotional expression without coercive demands
• Healthy boundary-setting
• Legitimate marketing from known brands
• Real security notifications and 2FA codes
• Friends or family asking for money in a normal way
• Passive-aggressive tone without a coercive condition attached

Using the App

Dashboard

The main window shows three stats (messages scanned, threats detected, safe messages) and a list of all flagged messages. Click any flag to expand it and see the full analysis - detected tactics, explanation, and a suggested response.

Menu Bar

Sentinel lives in your Mac's menu bar with a shield icon. Click it for quick access:

• Open Sentinel - brings the dashboard window back if you closed it
• Protection: ON/OFF - toggle scanning
• Stats - see scan and flag counts
• Quit - stop Sentinel

Notifications

When a threat is detected, Sentinel sends a macOS notification. Critical threats also trigger a prominent alert sound. Click the notification to open the dashboard and see the full analysis.

Protection Toggle

You can pause and resume scanning at any time from the dashboard toggle or the menu bar. When paused, the menu bar icon shows a sleep indicator.

Privacy

Sentinel was designed with privacy as the core requirement.

• 100% on-device - the AI model runs locally on your Mac's processor
• No cloud - no messages are ever uploaded to any server
• No internet required - after initial model download, Sentinel works fully offline
• No accounts - no sign-up, no login, no tracking
• No data collection - Sentinel doesn't collect analytics, usage data, or any information about you
• Your messages are not used for training - the AI model comes pre-trained and never learns from your data

FAQ

Does it work with SMS or just iMessage?

Sentinel reads from the macOS Messages database, which includes both iMessage and SMS texts received on your Mac.

Does it read messages I send?

No. Sentinel only analyzes incoming messages from other people.

Can I use it on iPhone?

Not currently. Apple doesn't allow third-party apps to access messages on iOS. Sentinel works on Mac where the Messages database is accessible with proper permissions.

How much RAM does it use?

About 4 GB while running. The AI model stays loaded in memory for fast analysis.

What if it flags something incorrectly?

No detection system is perfect. Sentinel achieves 90% precision, meaning roughly 1 in 10 flags may be a false positive. Use flagged messages as a prompt to think critically, not as a definitive judgment.

Is it free?

Sentinel is free during the beta period.

Troubleshooting

Sentinel isn't detecting any messages

• Make sure Full Disk Access is granted (see setup steps above)
• Make sure Protection is toggled ON in the dashboard
• Sentinel only detects messages that arrive after it starts running
• Try quitting and reopening Sentinel

"Unidentified developer" warning on launch

During beta, Sentinel isn't code-signed with Apple. To open it: right-click the app > Open > click "Open" in the dialog. You only need to do this once.

The AI model is taking a long time to download

The initial model download is about 700 MB. On a slow connection this can take a few minutes. After the first download, Sentinel works fully offline.