A fully working personal AI assistant you can run in your own home today.
While enterprise AI frameworks are blueprints, Personal Neuro Layer is a complete, running system built for real daily life — home automation, email, crypto trading, security monitoring, documents, memory, and voice control, all connected through a single intelligent routing layer.
| Module | Capability |
|---|---|
| 🧠 Neuro Layer | Central router — classifies intent and dispatches to the right organ |
| 💬 LLM (Local) | Powered by any LM Studio model (default: Qwen 2.5 14B) |
| 🏠 Home Assistant | Lights, switches, media, climate, scenes, sensors |
| 🎵 Music | Play any artist or album via Music Assistant |
| Read, search, compose, send via IMAP/SMTP | |
| 📄 Documents | Generate invoices, estimates, and letters as PDF |
| 📇 Contacts | Full contact management |
| 🔐 Security | Login protection, IP blacklisting, auto-ban via MikroTik SSH |
| 🌐 Network | MikroTik router stats, VPN, connected devices |
| 💾 NAS | QNAP storage management |
| 💰 Crypto | ADA paper + live trading, EMA crossover signals, Blockfrost wallet |
| 🌤️ Weather | Current conditions + forecast (Open-Meteo, no API key needed) |
| 🔍 Web Search | DuckDuckGo search + page fetch |
| 📰 News | Latest headlines |
| 🧠 Memory | Long-term JSONL memory + vector DB semantic search |
| 🎙️ Voice | Wake word listener + Tap to Talk (browser mic via HTTPS) |
| 🤖 Self-Review | Jarvis can review and suggest improvements to his own code |
User Input (text or voice)
↓
mcp_router_hub.py ← The Neuro Layer (central brain)
↓
Intent Classification
↓
┌──────────────────────────────────────────┐
│ LLM │ Memory │ Home Assistant │ Email │
│ Crypto │ Network │ Weather │ Search │ NAS │
└──────────────────────────────────────────┘
↓
Response → TTS → Audio
Each organ is isolated — add, remove, or swap any module without touching the others.
- Python 3.10+
- Windows (Linux/Mac support in progress)
- LM Studio — Download here — run any local LLM
- Home Assistant (optional) — for smart home control
- MikroTik router (optional) — for network monitoring and security
- QNAP NAS (optional) — for storage management
- Blockfrost account (optional) — for crypto wallet features
Jarvis runs its AI brain locally through LM Studio — performance depends on your hardware.
| Spec | Minimum | Recommended |
|---|---|---|
| RAM | 16GB | 32GB |
| GPU VRAM | 6GB | 8GB+ |
| CPU | 6-core modern | 8-core+ |
| Storage | 20GB free | 40GB+ free |
The default model is Qwen2.5-14B — a strong balance of intelligence and speed on mid-range hardware.
If your machine struggles, swap to a smaller 7B model in LM Studio with no other changes required. The model name can be updated in model_injection/cognition.py.
No GPU? CPU-only mode works but responses will be slow. A 7B model on a modern CPU is usable for light tasks.
git clone https://github.com/your-username/jarvis.git
cd jarvissetup.batEdit the config files created during setup:
Required:
mcp_servers_hub/login_security/auth_config.py— set your login username + password
Optional (enables each module):
mcp_servers_hub/home_assistant_server/ha_config.py— Home Assistant IP + tokenmcp_servers_hub/email_server/.env— email credentialsmcp_servers_hub/mikrotik_server/mikrotik_config.py— router IP + credentialsmcp_servers_hub/qnap_server/qnap_config.py— NAS IP + tokenmcp_servers_hub/crypto_wallet_server/wallet_config.py— Blockfrost project ID
Set your default weather location in mcp_servers_hub/internet_server/server.py:
DEFAULT_LOCATION = "Your City, ST"- Download and install LM Studio
- Load a model (recommended:
qwen2.5-14b-instruct) - Start the local server on port
1234
python app.pyhttp://localhost:5000
Log in with the credentials you set in auth_config.py.
To access Jarvis from your phone on your local network:
- Find your PC's local IP address (e.g.
192.168.1.100) - Open
http://192.168.1.100:5000in your phone's browser
For HTTPS (required for microphone access on mobile):
- Use a Cloudflare Tunnel to expose Jarvis securely over HTTPS
- Or run
python generate_cert.pyto create a self-signed certificate for LAN use
Wake Word (PC): Say "Hey Jarvis" — the PC microphone listens continuously when Jarvis is running.
Tap to Talk (Browser): Click the microphone button on the chat page — works on any device with a microphone over HTTPS.
Jarvis remembers things you tell him using a two-layer memory system:
- JSONL Memory — structured entries for facts, contacts, dates, preferences
- Vector DB — semantic search for natural language recall
Tell him anything:
"Remember that my wife's birthday is March 15" "Note that I prefer dark roast coffee"
He'll recall it in future conversations.
The trading module runs paper trades (simulated) in parallel with live trades at all times.
- EMA 5/20 crossover signals on 15-minute ADA candles
- Every live trade requires manual approval before execution
- Paper trading tracks performance silently — evaluate signals before going live
- Powered by Blockfrost + Minswap DEX
jarvis/
├── app.py # Flask boot layer
├── jarvis_routes.py # Web routes (chat, voice, pages)
├── model_injection/
│ ├── cognition.py # LLM call + memory injection
│ └── prompts.py # System prompt + context builder
├── mcp_servers_hub/
│ ├── mcp_router_hub.py # The Neuro Layer — central router
│ ├── home_assistant_server/ # Smart home organ
│ ├── email_server/ # Email organ
│ ├── crypto_wallet_server/ # Crypto trading organ
│ ├── mikrotik_server/ # Router organ
│ ├── qnap_server/ # NAS organ
│ ├── internet_server/ # Weather, search, news
│ ├── login_security/ # Auth + security scanner
│ ├── stt_server/ # Speech-to-text + wake word
│ ├── self_writing_server/ # Code review organ
│ └── memory_servers/ # JSONL + vector DB memory
├── templates/ # HTML UI pages
├── static/ # CSS, JS, images
└── setup.bat # Windows setup script
- React Native mobile app (iOS + Android)
- Google Tools organ (Calendar, Drive, Docs)
- Son's electrical business portal
- Linux/Mac setup script
- Docker container
MIT License — free to use, modify, and distribute.
Built with purpose. Engineered by Ekimtech.com