Civilization Node

Offline Knowledge Retrieval & LLM Interface

civilization_node is a self-hosted, offline-first application that integrates a local LLM inference engine with large-scale static knowledge bases (Wikipedia, StackOverflow, iFixit).

It leverages Retrieval Augmented Generation (RAG) to ground LLM responses in verifiable data stored in local ZIM archives, enabling high-fidelity technical assistance without an internet connection.

Designed for air-gapped environments, digital preservation, and off-grid reliability.

Architecture

The system is composed of three primary components:

• Inference Engine: Ollama running dolphin-llama3 (customized via Modelfile) for query processing and reasoning.
• Knowledge Base: Kiwix tools reading compressed ZIM archives (e.g., wikipedia_en_all_nopic.zim).
• Interface: Open WebUI serving as the frontend and orchestration layer.

System Requirements

Component	Specification
OS	Linux (Ubuntu 22.04+ recommended) or WSL2 (Windows)
RAM	12GB Minimum (16GB+ Recommended to prevent OOM kills)
Storage	100GB+ (Dependent on selected ZIM archives)
GPU	Optional. CPU inference works but is slower.

Installation

1. Dependencies

Ensure docker and docker compose are installed.

# Verify installation
docker compose version

2. LLM Backend Setup (Ollama)

Install Ollama and configure it to bind to 0.0.0.0 so the Docker container can access it.

# Install
curl -fsSL https://ollama.com/install.sh | sh

# Configure systemd override for network binding
sudo mkdir -p /etc/systemd/system/ollama.service.d
echo '[Service]
Environment="OLLAMA_HOST=0.0.0.0"' | sudo tee /etc/systemd/system/ollama.service.d/override.conf

# Reload and restart
sudo systemctl daemon-reload
sudo systemctl restart ollama

3. Model Provisioning

Pull the base model and create the custom system-prompt configuration.

# Pull base model
ollama pull dolphin-llama3

# Build the custom "Librarian" model
ollama create survival-librarian -f Modelfile.survival

# Pull embedding model for RAG (optional usage)
ollama pull nomic-embed-text

4. Application Deployment

Clone the repository and initialize the environment.

# Initialize directories and environment variables
cp .env.example .env
./setup_env.sh

# Start the application stack
docker compose up -d

Data Management

The system relies on ZIM archives. Use the included utility script to download them ensuring they are placed in the correct volume path.

./maintenance/list_available_content.sh

Recommended Archives:

• wikipedia_en_all_nopic: General encyclopedia.
• ifixit_en_all: Repair manuals.
• stackoverflow.com_en_all: Technical Q&A.

Configuration

Activating the Kiwix Tool

The LLM requires the Python tool definition to query the ZIM files.

1. Navigate to http://localhost:3000.
2. Go to Workspace > Tools > Create Tool.
3. Copy the content of kiwix_tool.py into the editor.
4. Save as "Kiwix".
5. Enable the tool in your Chat Settings when starting a new session.

Troubleshooting

Connection Refused: Ensure Ollama is running and listening on all interfaces. Check status:

systemctl status ollama
netstat -tulpn | grep 11434

ZIM Read Errors: Corrupt downloads can crash the reader. Verify checksums or delete the latest file in /opt/civilization/library/zims.