AIEGIS Documentation
User Manual
Complete guide to deploying and using the AIEGIS endpoint AI visibility platform — for IT administrators and compliance teams.
Version
v0.5
Platform
macOS · Windows · Browser
Updated
May 2026
Contents
Chapter 01

What is AIEGIS

AIEGIS is an endpoint AI visibility platform. It watches every interaction your employees have with AI tools — ChatGPT, Claude, Gemini, Copilot, Cursor — and gives your IT department a real-time view of what's happening, without the data ever leaving your infrastructure.

Why it exists

When an employee pastes a customer list into ChatGPT to write outreach emails, your organisation has no record of it. That data is now in OpenAI's systems. Under the EU AI Act, that is a compliance failure. Under your data protection obligations, it may be a breach.

AIEGIS closes that gap. Every prompt sent to an AI tool is intercepted before it leaves the device, checked against your organisation's policy, and logged to your own infrastructure. If a prompt violates policy, it is blocked before it reaches the AI. If it is allowed, it is logged with full context.

What AIEGIS is not

AIEGIS is not a cloud logging service. Your data never goes to AIEGIS Ltd servers. It is not a content filter that reads personal emails or browsing history — it only activates on known AI platforms. It is not a keylogger — it reads only prompt content submitted to AI tools.

Note
AIEGIS is scoped exclusively to AI platform domains: chatgpt.com, claude.ai, gemini.google.com, copilot.microsoft.com, and cursor.sh. It has no visibility into any other website or application.
Chapter 02

Before You Begin

AIEGIS has three components. You do not need to install all of them — choose the ones that match your deployment needs.

System requirements

ComponentRequirements
Browser ExtensionChrome, Edge, or Firefox · macOS, Windows, or Linux
Mac AgentmacOS 13 Ventura or later · Apple Silicon or Intel
API IntegrationAny platform that can make HTTPS POST requests

What you need from AIEGIS

Before deployment, your IT administrator should have the following ready:

01
API Key
Provided by AIEGIS after onboarding. Stored securely — treat it like a password. One key can cover all endpoints in your organisation.
02
Endpoint URL
Your AIEGIS endpoint: https://aiegis.ie for cloud deployment, or your self-hosted URL for on-premises.
03
Agent ID
A label identifying your organisation's deployment. Used in the dashboard to filter by department or location. Example: company-london-it
04
Policy decision
Should violations be blocked (prompt never reaches AI) or warned (prompt goes through, logged for IT review)? Default is block-on-violation.
Chapter 03

Browser Extension

The browser extension is the primary deployment method. It installs in Chrome, Edge, or Firefox and monitors all AI tool usage in the browser — no changes to employee workflow required.

How to install (IT administrator)

01
Obtain the extension package
Contact AIEGIS to receive the extension files. For enterprise deployment, we provide a signed CRX package suitable for Chrome Enterprise policy push.
02
Deploy via Chrome Enterprise (recommended)
Add the extension to your Chrome Enterprise admin console under Apps & Extensions → Force Install. Employees do not need to do anything — the extension installs automatically on next browser launch.
03
Load unpacked (for testing)
Open Chrome → go to chrome://extensions → enable Developer mode → click Load unpacked → select the aegis-browser-ext folder.
04
Configure the extension
Click the AIEGIS icon in the Chrome toolbar → enter your API Key, Endpoint URL, and Agent ID → save. Configuration is stored per-browser profile.

What employees see

When a prompt is allowed, nothing changes. The AI responds normally. When a prompt is blocked, the AI tool displays an error message explaining the prompt was blocked by organisational policy. Employees are not shown the specific policy rule triggered — only that a violation occurred.

Tip for IT
You can toggle enforcement mode from Block to Warn in the extension settings. Warn mode lets all prompts through but logs violations to the dashboard — useful during a pilot period when you want visibility before enforcing policy.

What is monitored

The extension intercepts the fetch() and XMLHttpRequest calls made by AI platforms. This gives it access to the prompt content before it leaves the device. It does not read browser history, cookies, passwords, or any content on non-AI websites.

Chapter 04

Mac Agent

The Mac agent provides deeper coverage for macOS users — including native app AI interactions that the browser extension cannot see, such as Copilot in Microsoft Word or AI features in Xcode.

Installation

01
Download the agent
Contact AIEGIS to receive the signed macOS application package (AiegisAgent.pkg).
02
Install system-wide
Double-click the .pkg file and follow the installer. The agent requires Accessibility permissions — macOS will prompt you to grant these in System Settings → Privacy & Security → Accessibility.
03
Configure credentials
Run aegis-agent --configure in Terminal. Enter your API key, endpoint URL, and agent ID when prompted. Credentials are stored securely in the macOS Keychain.
04
Verify it's running
Run aegis-agent --status in Terminal. A healthy response shows active: true and the current policy mode.

What the Mac agent monitors

AI SurfaceCovered
ChatGPT in browser✓ Browser extension + Mac agent
Claude in browser✓ Browser extension + Mac agent
Copilot in Microsoft 365✓ Mac agent (native app)
Cursor (AI code editor)✓ Mac agent
AI in Xcode✓ Mac agent
Personal device (unmanaged)✗ Not covered — policy only
Important
The Mac agent requires macOS Accessibility permissions. If your organisation uses MDM (Jamf, Mosyle, or similar), you can pre-grant these permissions via configuration profile, avoiding the per-user approval step.
Chapter 05

How Enforcement Works

Every prompt submitted to an AI tool passes through 15 security layers before a decision is made. This happens in under 200 milliseconds — fast enough that employees do not experience any perceptible delay on clean traffic.

The three decisions

Allow
Prompt passes all 12 enforced layers. Sent to the AI normally. Logged to your dashboard with full context.
Warn
Prompt triggers a policy rule but enforcement is set to Warn. Prompt goes through. Flagged in the IT dashboard for review.
Block
Prompt violates policy. Intercepted before reaching the AI. Employee sees a policy violation message. Logged with threat detail.

What triggers a block

AIEGIS blocks prompts that match your organisation's configured policy rules. Default rules cover:

PII and sensitive data
Email addresses, phone numbers, national ID numbers, credit card numbers, API keys, AWS credentials detected in the prompt body.
Prompt injection attempts
Patterns matching known jailbreak attempts, "ignore previous instructions" style inputs, and other injection signatures — 8 patterns checked.
Custom organisational rules
Rules your IT administrator has configured — competitor names, internal project codenames, regulated data categories.

Fail-open behaviour

If AIEGIS cannot reach the enforcement endpoint (network outage, maintenance), it fails open — prompts go through and are logged locally for later upload. This is a deliberate design decision: employee work is never blocked by an infrastructure failure outside your control.

Note
Fail-open can be changed to fail-closed for high-security deployments. Contact your AIEGIS administrator to enable strict mode.
Chapter 06

The IT Dashboard

The IT Violation Centre gives your security team a real-time view of every blocked and warned prompt across your organisation. See who triggered what, take action, and build your audit trail.

The dashboard is available at https://aiegis.ie/it/violations?key=YOUR_ADMIN_KEY. Your admin key is provided during onboarding — treat it as a privileged credential.

For a complete guide to every feature in the dashboard, see the Dashboard Manual →

Quick actions

ActionWhat it does
NotifyLogs that the employee was notified. Does not send a message automatically.
NoteAdds an investigator note to the violation record. Visible in audit exports.
EscalateLogs escalation to incident response. Adds the record to your escalation queue.
TrainingLogs a training recommendation for the employee.
DoneMarks the violation as reviewed. Moves it out of the pending queue.
Chapter 07

Privacy & Data

AIEGIS is designed to give organisations visibility while respecting employee privacy. Here is exactly what is captured, what is redacted, and where data goes.

What is captured

DataCaptured?Notes
Prompt textYesAfter PII redaction — emails, phones, ID numbers replaced with [REDACTED]
AI responseYesLogged after render, same redaction applied
Employee identityAgent ID onlyYou define the ID — we don't receive employee names
AI model usedYese.g. gpt-4o, claude-3-5-sonnet
TimestampYesUTC, millisecond precision
Passwords / tokensNeverRedacted before logging, not stored
Non-AI browsingNeverExtension scoped to AI domains only
TLS keysNeverNo TLS interception — sensor reads at application layer

Where data goes

In self-hosted deployments, all data goes directly from the endpoint to your infrastructure. AIEGIS Ltd never receives it. In cloud deployments, data is sent over TLS to your AIEGIS-managed instance — processed, then stored in the region of your choice. Default retention is 90 days, configurable.

EU AI Act & GDPR
Under EU AI Act Article 26.7, employees must be informed of AI monitoring. AIEGIS provides a worker notice template at /compliance/worker-notice-template — adapt it to your organisation's legal requirements before deployment.
Chapter 08

Troubleshooting

Extension not intercepting prompts

01
Check the extension is active
Go to chrome://extensions and confirm AIEGIS Endpoint AI Visibility is enabled and not paused.
02
Verify your API key
Click the AIEGIS toolbar icon. If the API key field is empty, the extension will fail-open — all prompts go through without enforcement.
03
Check the endpoint is reachable
Run curl -sI https://aiegis.ie/api/health. If you get a non-200 response, the enforcement endpoint is unreachable — check your network or firewall rules.

Mac agent not running

01
Check status
Run aegis-agent --status in Terminal. If it returns an error, the agent is not running.
02
Check Accessibility permissions
Go to System Settings → Privacy & Security → Accessibility. AIEGIS Agent must be listed and enabled.
03
Restart the agent
Run aegis-agent --restart. If the issue persists, reinstall the package and re-grant permissions.

Violations not appearing in dashboard

The dashboard auto-refreshes every 30 seconds. If violations are not appearing after 60 seconds, check: (1) your admin key in the URL is correct, (2) the agent ID in the extension matches what your dashboard is filtered to, (3) the endpoint and API key in the extension are correctly set.

Chapter 09

Technical Reference

API endpoints used by the extension

EndpointPurposeAuth
POST /api/protectInline enforcement check on every promptX-API-Key
GET /api/healthLiveness check on extension startupNone

Extension storage keys

KeyValueDefault
aegis_api_keyYour organisation's API keyEmpty (enforcement disabled)
aegis_endpointAIEGIS endpoint URLhttps://aiegis.ie
aegis_agent_idYour deployment agent IDbrowser-ext-agent
aegis_enforceBlock mode (true) or warn mode (false)true

Performance specifications

MetricValue
Enforcement latency p95<200ms (boundary)
CPU overhead (Mac agent, idle)<2%
Memory footprint (Mac agent)<50MB
Network overhead per prompt<100kbps average
Security layers checked per request15