Browser Extension
Browser extension that protects your AI conversations from data harvesting, monitors AI agents, controls MCP tools, and keeps you safe online.
Features
Core Protection
| Feature | Description |
|---|---|
| Conversation Shield | Blocks other extensions from reading your AI chats |
| Secret Scanner | Detects passwords, API keys, seed phrases before you send them |
| Extension Trust Score | Rates installed extensions by security risk |
| Real-time Alerts | Notifications when threats are detected |
Agent Shield
| Feature | Description |
|---|---|
| Agent Registry | Track all connected AI agents (ElizaOS, AutoGPT, CrewAI, custom) |
| Action Interceptor | Review and approve agent actions before execution |
| Memory Scanner | Detect memory injection attacks (40+ patterns) |
| Trust Levels | Assign trust levels for automatic approval |
MCP Gateway
| Feature | Description |
|---|---|
| Server Registry | Track all connected MCP servers |
| Tool Interceptor | Review and approve MCP tool calls |
| Risk Calculator | Automatic risk assessment for tool calls |
| Tool Validation | THSP-based validation for all tool executions |
Approval System
| Feature | Description |
|---|---|
| Configurable Rules | Auto-approve, auto-reject, or manual approval |
| Risk-based Defaults | Low risk auto-approved, high risk requires approval |
| Action History | Full audit trail of all actions |
| Expiration Handling | Automatic handling of expired requests |
Supported Platforms
| Platform | URL |
|---|---|
| ChatGPT | chat.openai.com, chatgpt.com |
| Claude | claude.ai |
| Gemini | gemini.google.com |
| Perplexity | perplexity.ai |
| DeepSeek | deepseek.com |
| Grok | grok.x.ai |
| Copilot | copilot.microsoft.com |
| Meta AI | meta.ai |
Installation
Chrome Web Store (Coming Soon)
1. Visit the Chrome Web Store
2. Search for "Sentinel Guard"
3. Click "Add to Chrome"
Development Build
# Clone the repository
git clone https://github.com/sentinel-seed/sentinel.git
cd sentinel/packages/browser
# Install dependencies
npm install
# Build extension
npm run build
# The built extension will be in ./distLoad in Chrome
1. Open Chrome and go to chrome://extensions
2. Enable "Developer mode" (top right)
3. Click "Load unpacked"
4. Select the dist folder
Load in Firefox
1. Open Firefox and go to about:debugging#/runtime/this-firefox
2. Click "Load Temporary Add-on"
3. Select manifest.json from the dist folder
Load in Edge
1. Open Edge and go to edge://extensions
2. Enable "Developer mode"
3. Click "Load unpacked"
4. Select the dist folder
How It Works
Conversation Shield
The extension injects a content script into AI chat platforms that:
| Function | Description |
|---|---|
| Monitor | Detects suspicious DOM access attempts |
| Protect | Creates protective wrappers around conversation elements |
| Block | Prevents unauthorized data extraction |
Secret Scanner
Based on the same patterns used in the Sentinel VS Code extension:
| Capability | Description |
|---|---|
| Real-time scanning | Scans text as you type |
| 30+ secret types | API keys, passwords, tokens, seed phrases |
| Pre-submission warning | Options to remove, mask, or proceed |
The THSP Protocol
All actions are validated through the Truth-Harm-Scope-Purpose protocol:
| Gate | Question |
|---|---|
| Truth | Is this from a legitimate source? |
| Harm | Could this cause harm? |
| Scope | Is this within appropriate boundaries? |
| Purpose | Is there a legitimate reason? |
Agent Shield
Monitors AI agents and their actions:
| Function | Description |
|---|---|
| Intercept | Catches tool calls before execution |
| Detect | Identifies memory injection attempts |
| Approve | Requires approval for high-risk actions |
MCP Gateway
Controls Model Context Protocol tool calls:
| Function | Description |
|---|---|
| Validate | Tools validated against THSP protocol |
| Calculate | Risk level assessment for each call |
| Audit | Maintains complete audit trail |
Internationalization
Sentinel Guard supports multiple languages:
| Language | Code |
|---|---|
| English | en |
| Spanish | es |
| Portuguese | pt |
Language is auto-detected from browser settings or can be set manually in Settings.
Privacy
Sentinel Guard is privacy-first:
| Principle | Description |
|---|---|
| No data collection | Your conversations never leave your browser |
| No external servers | All processing happens locally |
| Open source | Verify the code yourself |
Development
# Install dependencies
npm install
# Development build with watch
npm run dev
# Production build
npm run build
# Run tests
npm test
# Run tests with coverage
npm run test:coverage
# Lint code
npm run lintTesting
# Run all tests
npm test
# Run with coverage
npm run test:coverage
# Watch mode
npm run test:watchCurrent test coverage: 866 tests passing, 55%+ code coverage.
Part of Sentinel Ecosystem
| Component | Description |
|---|---|
| VS Code Extension | Protect your code |
| JetBrains Plugin | IDE integration |
| Neovim Plugin | Editor integration |
| Python/npm SDKs | For applications |
| Browser Extension | Protect your browsing |