en Deployment

SelenaCore Deployment and Installation Guide

This guide covers hardware requirements, installation, configuration, and ongoing operations for the SelenaCore smart home hub.

---

Supported Hardware

Platform	Notes
Raspberry Pi 4/5	4GB+ RAM recommended
NVIDIA Jetson Orin Nano	GPU-accelerated TTS/STT support
Any Linux SBC (ARM64 or x86_64)	Tested on Ubuntu and Debian-based distros

Minimum requirements:

• 2GB RAM — sufficient for core functionality without a local LLM
• 4GB+ RAM — required for full features including Ollama-based local LLM inference

---

OS and Software Requirements

• Ubuntu 22.04+ or Raspberry Pi OS (Bookworm)
• Docker 24+ and Docker Compose v2
• Python 3.11+

---

Installation

Unified Installer (Recommended)

git clone https://github.com/dotradepro/SelenaCore.git
cd SelenaCore
sudo ./install.sh

install.sh is the only script users run. It does the bare minimum to get the system reachable in a browser, then prints a URL like http://<lan-ip>/. Open that URL — the rest of the setup happens inside the first-run wizard (model selection, downloads, admin user, platform registration, native systemd services) with a live progress bar.

What install.sh does:

1. Detects hardware (Jetson / Raspberry / CUDA / generic Linux)
2. apt-get install host packages (Docker, FFmpeg, arp-scan, pulseaudio, nmcli, …)
3. Creates the selena system user and adds it to docker/audio/video groups
4. Creates /var/lib/selena/{models,…}, /var/log/selena, /secure
5. Seeds Piper voices from ~/.local/share/piper/models/ if present
6. Copies config/core.yaml.example → config/core.yaml with wizard.completed=false
7. npx vite build (frontend bundle)
8. docker compose up -d --build (selena-core + selena-agent)
9. Stages smarthome-core.service / smarthome-agent.service into /etc/systemd/system/ (not enabled — the wizard does that)
10. Prints a banner with the wizard URL

install.sh does NOT download Whisper / Vosk / Piper voices / Ollama models — those are downloaded by the wizard with progress visible in the browser.

Manual Setup

# Clone the repository
git clone https://github.com/dotradepro/SelenaCore.git
cd SelenaCore

# Copy and edit the environment file
cp .env.example .env
# Edit .env with your settings (see Environment Variables below)

# Build and start
docker compose build
docker compose up -d

---

Docker Architecture

The docker-compose.yml file defines two services.

selena-core (main service)

The primary container running the SelenaCore application.

• Image: Built from Dockerfile.core (base: python:3.11-slim)
• Network mode: host (required for audio and device access)
• Privileged: true (required for hardware access)
• Exposed ports:
  • 80 — Unified API + Web UI (single process)
  • 443 — HTTPS (TLS proxy to :80)
• Volumes:
  • /var/run/docker.sock — Docker socket for managing module containers
  • selena_data:/var/lib/selena — Database, voice models, backups
  • selena_secure:/secure — Encrypted tokens and keys
  • /dev/snd — ALSA sound devices for audio input/output
  • Ollama models directory (if configured)
• Health check: GET /api/v1/health every 30 seconds
• Bundled software: FFmpeg, PortAudio, VLC, ALSA utils (aplay, arecord, amixer)
• External services (native on host): Piper TTS (piper-tts.service), Ollama

selena-agent (integrity agent)

A separate container that continuously monitors core integrity.

• Performs SHA256 hash verification of core files every 30 seconds
• On integrity violation: stops modules, sends notification, initiates rollback, enters SAFE MODE

GPU Support (NVIDIA Jetson)

For GPU-accelerated container features, start with the GPU override file:

docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d

Piper TTS Native Service

Piper TTS runs natively on the host (not in Docker) for direct GPU access and lower memory usage.

# Install Piper TTS
pip3 install --user piper-tts aiohttp

# For GPU on Jetson (JetPack 6, CUDA 12.x):
pip3 install --user onnxruntime-gpu --extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu126
pip3 install --user "numpy<2"
sudo ln -sf /usr/lib/aarch64-linux-gnu/libcudnn.so.9 /usr/lib/aarch64-linux-gnu/libcudnn.so
# Or use the automated script: bash scripts/build-onnxruntime-gpu.sh

# Deploy systemd service
sudo cp scripts/piper-tts.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable --now piper-tts

# Verify
curl http://localhost:5100/health
# → "device": "gpu", "cuda_available": true

Note: PyPI onnxruntime-gpu does NOT have aarch64 wheels. Must use NVIDIA Jetson AI Lab index.

Device modes: --device auto (default, detect GPU), --device cpu, --device gpu

Ollama

Local LLM inference runs natively on the host via Ollama for GPU acceleration and to keep the container slim. SelenaCore communicates with Ollama over its HTTP API (OLLAMA_URL, default http://localhost:11434).

# Install Ollama (one-time)
curl -fsSL https://ollama.com/install.sh | sh

# Enable and start the systemd service
sudo systemctl enable --now ollama

# Pull a model
ollama pull qwen2.5:3b

# Verify
curl http://localhost:11434/api/tags

Cloud LLM providers (OpenAI, Anthropic, Google AI, Groq) are configured via the voice settings UI and require no host-side service.

---

Environment Variables

All configuration is managed through the .env file in the project root. Copy .env.example to .env and adjust as needed.

Variable	Default	Description
CORE_PORT	80	API server port
CORE_DATA_DIR	/var/lib/selena	Data directory (DB, models)
CORE_SECURE_DIR	/secure	Encrypted secrets directory
CORE_LOG_LEVEL	INFO	Log level
DEBUG	false	Enable debug mode and Swagger UI
PLATFORM_API_URL	https://selenehome.tech/api/v1	Cloud platform URL
PLATFORM_DEVICE_HASH	(empty)	Device identification hash
UI_PORT	80	Web UI port
UI_HTTPS	true	Enable HTTPS for UI
DOCKER_SOCKET	/var/run/docker.sock	Docker socket path
MODULE_CONTAINER_IMAGE	smarthome-modules:latest	User module container image
GOOGLE_CLIENT_ID	(empty)	Google OAuth client ID
GOOGLE_CLIENT_SECRET	(empty)	Google OAuth secret
TUYA_CLIENT_ID	(empty)	Tuya integration client ID
TUYA_CLIENT_SECRET	(empty)	Tuya integration secret
TAILSCALE_AUTH_KEY	(empty)	Tailscale VPN auth key
GEMINI_API_KEY	(empty)	Cloud LLM fallback key
DEV_MODULE_TOKEN	(empty)	Development token for testing
OLLAMA_MODELS_DIR	(empty)	Ollama model storage directory

---

core.yaml Configuration

The main configuration file is located at /opt/selena-core/config/core.yaml. See Configuration Reference for all available options.

---

Systemd Services

To run SelenaCore as a system service that starts on boot, install the following unit file.

smarthome-core.service

# /etc/systemd/system/smarthome-core.service
[Unit]
Description=SelenaCore Smart Home Hub
After=docker.service
Requires=docker.service

[Service]
WorkingDirectory=/opt/selena-core
ExecStart=/usr/bin/docker compose up
ExecStop=/usr/bin/docker compose down
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

Enable and start:

sudo systemctl daemon-reload
sudo systemctl enable smarthome-core.service
sudo systemctl start smarthome-core.service

Additional Services

Service	Purpose
smarthome-agent.service	Integrity monitoring agent
piper-tts.service	Native Piper TTS HTTP server (installed only when voice.tts.primary.cuda: true)
selena-display.service	Wayland kiosk display (installed automatically when cage + connected DRM output detected)
vosk-server.service	Vosk STT server
piper-tts.service	Piper TTS server (native, GPU)

Headless Kiosk (Recommended for Production)

For Jetson and Raspberry Pi deployments, disable the desktop environment and run Chromium via Xorg kiosk. This saves ~1 GB RAM.

# Disable desktop, enable kiosk
sudo systemctl set-default multi-user.target
sudo systemctl disable gdm3

See Kiosk Setup for full instructions including getty autologin, ALSA audio, and Xorg configuration.

---

Onboarding Wizard

On first start, SelenaCore enters setup mode and walks the user through initial configuration.

1. Creates a WiFi access point: SmartHome-Setup
2. Opens a web wizard at http://192.168.4.1
3. Wizard steps:
   • Language selection
   • WiFi network configuration
   • Device name
   • Voice engine selection
   • User profile creation
   • Display settings
   • Platform link
4. After completion, the system restarts in normal mode

---

Operations

Health Check

Verify the system is running correctly:

curl http://localhost/api/v1/health

Expected response:

{
  "status": "ok",
  "version": "...",
  "mode": "normal",
  "uptime": 12345,
  "integrity": "ok"
}

Viewing Logs

# Follow core logs in real time
docker compose logs -f selena-core

# Filter logs for a specific module
docker compose logs -f selena-core | grep "module-name"

# View log files on disk
ls /var/log/selena/

Data Directories

Path	Contents
/var/lib/selena/	SQLite database, voice models, backups
/var/lib/selena/models/vosk/	Vosk STT models
/var/lib/selena/models/piper/	Piper TTS models
/secure/	Encrypted tokens, AES keys
/secure/module_tokens/	Module authentication tokens

---

Updating

Pull the latest changes and rebuild:

cd /opt/selena-core
git pull
docker compose build
docker compose up -d

Alternatively, use the update_manager system module for automatic over-the-air updates.

---

Backup

The backup_manager system module handles automated backups:

• Local backups: SQLite database and configuration files
• Cloud backups: To configured remote storage

For manual backup, copy the data and secrets directories:

sudo cp -r /var/lib/selena/ /path/to/backup/selena_data/
sudo cp -r /secure/ /path/to/backup/selena_secure/

---

Troubleshooting

Problem	Solution
Port 80 in use	Change CORE_PORT in .env and restart
No audio output or input	Check /dev/snd is mounted in docker-compose.yml; verify devices with aplay -l and arecord -l inside the container; use plughw:X,Y device IDs for ALSA
Module will not connect	Verify MODULE_TOKEN and SELENA_BUS_URL are set correctly in the module environment
System entered Safe Mode	Check integrity agent logs (docker compose logs selena-agent); verify core file hashes match expected values
Docker permission denied	Ensure the current user is in the docker group, or run with sudo
Ollama models not loading	Verify OLLAMA_MODELS_DIR points to an existing directory with sufficient disk space