AI content detection extension that analyzes text to determine if it was written by AI or human. Features popup analysis, right-click context menu, and floating popup on text selection.
- Instant Detection - Highlight text, get a score
- Floating Popup - Select text on any webpage, click "Check AI"
- Context Menu - Right-click selected text to analyze
- Popup Interface - Paste text directly into the extension
- Clone this repo
- Go to
chrome://extensions - Enable "Developer mode"
- Click "Load unpacked"
- Select this folder
The extension sends text to a detection API that runs multiple ML models:
- OpenAI's text classifier
- ChatGPT detector
- Fakespot model
- Ensemble scoring
Results come back with a "human score" (0-100%) and confidence level.
Three ways to use it:
- Floating popup - Select text on any page, a small popup appears. Click "Check AI".
- Right-click - Select text, right-click, choose "Check with AI Detector"
- Extension popup - Click the extension icon, paste text, hit detect
├── manifest.json # Extension config
├── content.js # Injected into pages, handles floating popup
├── content.css # Styles for floating popup
├── popup.html # Extension popup UI
├── popup.js # Extension popup logic
├── background.js # Service worker for context menu
└── icons/ # Extension icons
The extension hits app.parallellives.ai/api/detect for detection. This is a hosted API - you don't need to run anything locally.
- Free - Basic ML detection, daily limit
- Pro - Higher accuracy models, humanization tips, plagiarism check, no limits
Pro requires a subscription at parallellives.ai.
The extension is a standard Chrome Manifest V3 extension. Content script injects into all pages and listens for text selection. When detected, it shows a floating popup anchored near the selection.
The detection API runs on Vercel (Next.js). Nothing fancy - just API routes that call ML models and return scores.
For tracking and analytics, the backend uses TiDB (MySQL-compatible distributed database). Here's how it's set up:
Why TiDB:
- MySQL compatibility means standard queries, no learning curve
- Serverless tier is free and handles the scale I need
- Distributed by default, so I don't worry about scaling later
What's stored:
detector_usage- Every detection gets logged (text hash, score, site URL, timestamp)detector_cache- Results cached for 24 hours to avoid re-running ML models on same textdetector_feedback- When users click "accurate" or "wrong", it gets stored here
Schema example:
CREATETABLEdetector_usage (
id VARCHAR(36) PRIMARY KEY,
user_id VARCHAR(36),
text_hash VARCHAR(64) NOT NULL,
word_count INTNOT NULL,
score INTNOT NULL,
confidence VARCHAR(20) NOT NULL,
provider VARCHAR(50) NOT NULL,
site_url VARCHAR(500),
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);
CREATETABLEdetector_cache (
text_hash VARCHAR(64) PRIMARY KEY,
result JSON NOT NULL,
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
expires_at TIMESTAMPNOT NULL
);
CREATETABLEdetector_feedback (
id VARCHAR(36) PRIMARY KEY,
user_id VARCHAR(36),
text_hash VARCHAR(64) NOT NULL,
feedback_type ENUM('false_positive', 'false_negative', 'accurate') NOT NULL,
original_score INTNOT NULL,
comment TEXT,
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);The feedback loop is the interesting part. When enough users report a false positive on similar text patterns, I can use that data to tune the detection. Not doing anything fancy with it yet, but the data's there.
Connection is straightforward - TiDB Cloud gives you a MySQL connection string, and I use mysql2 in Node:
importmysqlfrom"mysql2/promise";constpool=mysql.createPool({host: process.env.TIDB_HOST,port: 4000,user: process.env.TIDB_USER,password: process.env.TIDB_PASSWORD,database: process.env.TIDB_DATABASE,ssl: {ca: fs.readFileSync("/etc/ssl/cert.pem")}});That's it. No ORM, no abstraction layers. Just SQL.
MIT