2.2M events/sec on ARM64 · 73.7% detection · 0% false positives · Zero-cloud · On-device
| Question | Answer |
|---|---|
| What is it? | Sovereign cyber defense mapping 64 Vedic Sutras to modern AI |
| How fast? | 2.2M events/sec on ARM64 (Android/Termux) |
| How accurate? | 73.7% phishing detection with 0% false positives |
| Where does data go? | Nowhere. All detection is on-device. Zero cloud. |
| What makes it unique? | Explainable Nyaya logic instead of black-box ML |
| Who is it for? | India's 1.4 billion. Runs on ₹8,000 phones. |
| Module | Vedic Sutra | Real Function |
|---|---|---|
| CHARAKA | Sutra 14 | Anomaly detection via Vata/Pitta/Kapha dosha classification |
| NYAYA | Sutra 3 | Phishing detection via 7-Pramana logical inference |
| GANDHARVA | Sutra 20 | Voice deepfake detection via harmonic fingerprinting |
| AKASHA | Sutra 45 | SHA-256 immutable forensic ledger |
| SUSHRUTA | Sutra 19 | Autonomous incident response (4 Mandukya states) |
| DHARMA | Sutra 30 | Shadow AI monitoring (5-layer Pancha Kosha analysis) |
| SARATHI | Sutra 39 | Unified orchestrator |
KAVACH doesn't use black-box ML. Every detection decision is explainable through 7 Pramanas (proofs):
| Pramana | Type | Weight | What It Detects |
|---|---|---|---|
| P₁ | Action/Urgency | 0.10 | "verify", "update", "click", "pay" |
| P₂ | Emotional Rasa | 0.25 | Fear, urgency, attraction, greed |
| P₃ | Authority Claim | 0.20 | "SBI", "HDFC", "PM Kisan", "APDCL" |
| P₄ | Sensitive Data | 0.30 | Aadhaar, PAN, KYC, OTP, passwords |
| P₅ | Suspicious URL | 0.15 | .tk, .ml, .xyz, .cc domains |
| P₆ | Temporal Urgency | 0.05 | "within 24 hours", "by tonight" |
| P₇ | Call-to-Action | 0.10 | Phone numbers + "call/dispute" |
| Metric | Value | Environment |
|---|---|---|
| Detection Rate | 73.7% | 19 Indian scam SMS samples |
| False Positives | 0% | 10 benign SMS samples |
| C++ Throughput | 2.2M events/sec | ARM64, 8 threads, Aho-Corasick |
| Python Throughput | 90K/sec | Single-threaded reference |
| Patterns | 124 | P1-P7 + Hinglish transliteration |
| Rule Index | 150K | Hash-indexed keyword engine |
📖 Full methodology: BENCHMARKS.md
KAVACH is DPDPA-compliant by architecture:
- ✅ All detection on-device — zero cloud
- ✅ Sensitive data (Aadhaar/PAN/OTP) processed in volatile RAM only
- ✅ Redacted in forensic logs
- ✅ No analytics, no telemetry, no external APIs
- ✅ MIT licensed — audit everything
📖 Full privacy policy: PRIVACY.md
Click the Open in Colab badge at top. No installation.
pkg install python git
git clone https://github.com/divineearthly/KAVACH-Sovereign-Intelligence.git
cd KAVACH-Sovereign-Intelligence
pip install -r requirements.txt
python kavach_main.pyC++ Engine (2.2M events/sec)
cd cpp-engine && mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release && make -j$(nproc)
./kavach🏗️ Architecture
KAVACH/
├── modules/ # Python defense modules
│ ├── charaka.py # Anomaly detection
│ ├── nyaya.py # Logic firewall (7-Pramana)
│ ├── gandharva.py # Voice analysis
│ ├── akasha.py # Immutable ledger
│ ├── sushruta.py # Incident response
│ └── dharma.py # Shadow AI monitor
├── cpp-engine/ # C++ high-performance engine
│ ├── include/kavach/ # Aho-Corasick, Sutra constants
│ ├── src/ # Detection pipeline
│ └── rules/ # 150K threat signatures
├── tests/ # Real-world phishing tests
├── api.py # FastAPI REST integration
└── kavach_main.py # SARATHI orchestrator
🌐 Live Ecosystem
Service Description ⚖️ Nyaya Lens API Vedic logical reasoning engine 🌾 Krishi-Veda Vedic agricultural AI 🤖 @NyayaLensBot Telegram bot 🧮 Vedic Math CNN Sutra-based CNN
🎯 Roadmap
Phase Milestone Status 1 Python modules (6/6) ✅ 2 C++ Aho-Corasick pipeline ✅ 3 124 Pramana patterns + Hinglish ✅ 4 MITRE ATT&CK mapping 📋 5 OpenSSF Best Practices badge 📋 6 IndiaAI Mission application 📋
📜 License
MIT © 2026 divineearthly
🕉️ सत्यमेव जयते — Truth Alone Triumphs 🔱
Built solo in Silchar, Assam · Android/Termux · Zero budget · Sovereign AI