Chrome extension (MV3) for real-time scam detection, built for India-specific threat patterns.
Version 6.2 — Chrome/Chromium only — Load unpacked (no Web Store listing yet)
Chetana scores every page you visit (0–100 trust) and scans for active fraud patterns before you interact. It runs a four-layer fallback chain so analysis works offline, on slow connections, and at scale:
domain cache (24h) → Ollama local (phi4-mini) → cloud API → local gate (10 rules, offline)
Key capabilities:
- Badge trust score (0–100) on every page
- WhatsApp Web live message scan
- Telegram Web forwarded-message scan
- Right-click context menu: "Check with Chetana"
- Side panel with full analysis
- Form guard — warns before submitting sensitive data to low-trust domains
- Link hover tooltips with pre-flight trust check
- Persistent domain cache (24h TTL)
- Clone the repo:
git clone https://github.com/MirrorDNA-Reflection-Protocol/chetana-browser cd chetana-browser - Open Chrome and navigate to
chrome://extensions/ - Enable Developer mode (top-right toggle)
- Click Load unpacked
- Select the
extension/directory inside the repo - Pin the Chetana icon from the Extensions menu
No build step required. The extension is plain MV3 — no bundler, no compile.
The local backend runs phi4-mini via Ollama for on-device deep analysis. Without it, the extension falls back to the cloud API or the offline local gate.
Requirements: Ollama installed, ~3GB disk for phi4-mini.
# 1. Create the model
ollama create chetana-guard-fast -f models/chetana-guard-fast.Modelfile
# 2. Start the FastAPI backend
cd backend
pip install -r requirements.txt
uvicorn main:app --host 127.0.0.1 --port 8799
# Verify
curl http://localhost:8799/healthBackend listens on localhost:8799. The extension detects it automatically — no config needed.
Browser Request
│
▼
┌─────────────────────┐
│ 1. Domain Cache │ Hit? Return cached score instantly (24h TTL)
└────────┬────────────┘
│ Miss
▼
┌─────────────────────┐
│ 2. Ollama Local │ phi4-mini at localhost:8799 — full page analysis
│ (FastAPI) │ Returns structured trust score + reason
└────────┬────────────┘
│ Unavailable
▼
┌─────────────────────┐
│ 3. Cloud API │ chetana.activemirror.ai/api/scan/full
│ (ActiveMirror) │ Requires internet, higher latency
└────────┬────────────┘
│ Unavailable
▼
┌─────────────────────┐
│ 4. Local Gate │ 10 India-specific regex/pattern rules
│ (offline) │ Zero latency, zero network dependency
└─────────────────────┘
Patterns tuned for Indian threat landscape:
| Category | Examples |
|---|---|
| UPI fraud | Fake payment links, QR code redirect scams |
| Digital arrest | Fake CBI/ED/police notices demanding payment |
| OTP theft | Fake OTP verification pages, SIM swap lures |
| Lottery / prize scams | "You won KBC/BSNL lottery" pages |
| Deepfake links | Video call scam landing pages |
| WhatsApp scams | Job offers, investment groups, forward chains |
| Telegram scams | Forwarded fraud messages, fake channel impersonation |
The offline local gate covers the most common 10 pattern classes without any network call.
chetana-browser/
├── extension/ # Chrome MV3 extension (load this)
│ ├── manifest.json
│ ├── background.js
│ ├── content.js
│ ├── sidepanel/
│ └── icons/
├── backend/ # FastAPI server (optional, port 8799)
│ ├── main.py
│ └── requirements.txt
├── models/
│ └── chetana-guard-fast.Modelfile # phi4-mini Ollama model def
└── README.md
Advisory tool. Scores are probabilistic — not legal determinations. Not affiliated with any government body, law enforcement agency, or financial regulator.
Built by ActiveMirror / N1 Intelligence OPC Pvt Ltd, Bengaluru/Goa, India. Part of the ActiveMirror / MirrorDNA ecosystem.
GitHub: https://github.com/MirrorDNA-Reflection-Protocol/chetana-browser
MIT License.