🤖 Bunker Bot — Offline AI Setup

Bunker Bot (Offline Survival AI) runs entirely on your local machine using Ollama — no internet, no API keys, no data leaving your device. This guide walks you through setup on any operating system.

✅

What You Need

A computer (macOS, Windows, or Linux), at least 8GB RAM (16GB recommended for larger models), and disk space for the model (4–40GB depending on model choice). Initial setup requires internet. After that, Bunker Bot works completely offline.

🦙

Easiest option: a llamafile (no install)

Prefer zero setup? A llamafile is a single file that is the AI model and the server — no Ollama, no install. Download one (e.g. Llama-3.2-3B, ~2.3 GB), make it executable, and run it:

chmod +x your-model.llamafile
./your-model.llamafile --server --nobrowser

Then open Bunker Bot, pick the 🦙 Llamafile mode (gear → Mode), and it connects to localhost:8080 automatically. The Ollama guide below is the alternative if you want a model manager. Get a llamafile from the Download page or Mozilla's llamafile project.

1. Install Ollama

Ollama is a free, open-source tool for running large language models locally. One installer, works on all platforms.

macOS (Apple Silicon & Intel)

Download the macOS installer from ollama.com/download — look for the macOS package (Ollama-darwin.zip).
Unzip the downloaded file. Move Ollama.app to your /Applications folder.
Double-click Ollama.app to launch it. A llama icon appears in your menu bar — this means it's running.
Open Terminal (Spotlight Search → "Terminal") and verify the installation:

ollama --version

You should see a version number like ollama version 0.x.x.

💡

Apple Silicon (M1/M2/M3/M4)

Ollama natively uses Apple Silicon's Neural Engine and unified memory for dramatically faster inference than CPU-only. A 7B model runs comfortably on a base M1 MacBook Air with 8GB unified memory.

Windows 10/11

Download OllamaSetup.exe from ollama.com/download.
Run the installer. Accept any Windows security prompts. Ollama installs automatically and starts as a background service.
Look for the Ollama icon in the System Tray (bottom right taskbar area, you may need to click the ^ arrow to show hidden icons).
Open Command Prompt or PowerShell and verify:

ollama --version

💡

NVIDIA GPU Support

Windows users with NVIDIA GPUs should install CUDA drivers for dramatically faster inference. Ollama detects and uses the GPU automatically. AMD GPUs with ROCm are also supported on Windows.

Linux (Ubuntu, Debian, Fedora, and others)

The official one-line install script:

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

This downloads and installs Ollama as a systemd service that starts automatically.

Verify installation: ollama --version
Check service status: systemctl status ollama
If not running: sudo systemctl start ollama
Enable auto-start on boot: sudo systemctl enable ollama

💡

GPU Support on Linux

NVIDIA: Install nvidia-container-toolkit for GPU acceleration. AMD: Ollama supports AMD ROCm on compatible GPUs (RX 6000 series and newer). Both are detected automatically.

2. Pull a Model

After Ollama is installed, download a language model. This requires internet and can take several minutes depending on model size and your connection speed.

# Recommended default — good balance of speed and capability
ollama pull llama3

# Or a specific size variant:
ollama pull llama3:8b     # 4.7GB — faster, less capable
ollama pull llama3:70b    # 40GB — slower, much more capable

# Alternative models:
ollama pull gemma3        # Google's Gemma 3
ollama pull phi3          # Microsoft's Phi-3 Mini — excellent for low-RAM systems
ollama pull mistral       # Mistral 7B — very fast and capable

After downloading, test the model:

ollama run llama3
# Type: "What are the top 5 priorities in a survival situation?"
# Type /bye to exit

List all installed models:

ollama list

3. Run Ollama as a Local Server

Bunker Bot connects to Ollama's REST API on localhost:11434. On macOS and Windows, the Ollama app runs the server automatically when it's in the menu bar / system tray. On Linux, the systemd service handles this.

Manual Server Start

# Start the server manually (if not auto-starting):
ollama serve

# Verify the server is running (test in a browser or curl):
curl http://localhost:11434/api/tags

You should see a JSON response listing your installed models.

CORS Configuration (Required for Browser Access)

By default, Ollama only accepts connections from localhost — which is what we want for security. The Bunker Bot chat panel in this guide sends requests to http://localhost:11434. If you see connection errors, verify Ollama is running and the port is not blocked by a firewall.

⚠️

Browser CORS Restriction

Some browsers block fetch() calls from a local file (file:///) to localhost. If Bunker Bot shows a connection error despite Ollama running, try: opening the guide via a local HTTP server (python3 -m http.server 8080 in the guide folder, then open http://localhost:8080). Or use Firefox, which has more permissive localhost policies.

4. Portable USB/SSD Setup

Store Ollama + models on a portable drive so you can run Bunker Bot on any compatible computer without downloading anything.

What You Need

USB-C SSD (portable, fast — USB-A will work but be slower): minimum 64GB for one model, 500GB+ for several
The Ollama installer for your target platform
The downloaded models

macOS Setup

Copy Ollama.app to the USB drive.
Move the models to a folder on the USB drive. Ollama stores models in ~/.ollama/models/ by default.

Set the OLLAMA_MODELS environment variable to point to the USB drive location before launching:

OLLAMA_MODELS=/Volumes/MyUSB/ollama-models /Volumes/MyUSB/Ollama.app/Contents/MacOS/ollama serve

Create a shell script on the USB drive (start-aria.sh) that sets this variable and launches Ollama automatically.

Windows Setup

Copy the Ollama installer to the USB drive.
After installing on the target computer, move the models folder to USB:

# Move model directory to USB drive
robocopy %USERPROFILE%\.ollama\models E:\OllamaModels /E /MOVE
# Set environment variable and restart Ollama
setx OLLAMA_MODELS "E:\OllamaModels"

Create a batch file (start-aria.bat) on the USB:

@echo off
set OLLAMA_MODELS=E:\OllamaModels
start "" "C:\Users\%USERNAME%\AppData\Local\Programs\Ollama\ollama.exe" serve

Linux Setup

Mount your USB drive and create a models directory on it.
Copy models: cp -r ~/.ollama/models /mnt/myusb/ollama-models

Launch Ollama pointing to USB models:

OLLAMA_MODELS=/mnt/myusb/ollama-models ollama serve &

✅

Speed Tip

USB-A 3.0 drives have 50–80 MB/s read speed — adequate but slow for large model files. A USB-C NVMe SSD enclosure (500+ MB/s) makes a significant difference in load times for large models. For field use, a 1TB Samsung T7 Shield or similar ruggedized portable SSD is recommended.

5. Model Selection by Hardware

Hardware Tier	RAM	Recommended Model	Size	Quality
Low (laptop, Raspberry Pi 5)	4–8GB	phi3:mini	2.3GB	Good for simple Q&A
Low	4–8GB	gemma:2b	1.4GB	Minimal, fast
Mid (modern laptop, Apple M1/M2)	8–16GB	llama3:8b	4.7GB	Very good — default recommendation
Mid	8–16GB	phi3:medium	7.9GB	Excellent for its size
Mid	16GB	mistral:7b	4.1GB	Fast and capable
High (workstation, gaming PC with GPU)	24GB+	llama3:70b	40GB	Excellent — near GPT-4 quality
High	32GB+	mixtral:8x7b	26GB	Very capable, mixture of experts
High (M2/M3 Ultra Mac)	64GB+	llama3:70b (quantized)	40GB	Best available offline model

💡

Quantization

Models come in different quantization levels (Q4, Q5, Q8, F16). Lower quantization = smaller file, faster speed, slightly lower quality. For survival use cases, Q4 or Q5 quantization is entirely adequate — the quality difference is negligible for practical questions. Ollama defaults to the optimal quantization for your hardware.

6. Fully Airgapped Operation

After initial setup (downloading Ollama and your model), Bunker Bot works with zero internet connectivity.

Setup Checklist for Airgapped Use

Ollama installed on target machine (or on USB drive)
Model downloaded and stored locally or on USB
This survival guide folder (all files) stored locally or on USB
Test: disconnect from internet → open guide → launch Bunker Bot → ask a question → verify response
Test on a secondary machine using only the USB drive (no internet on that machine)

What Works Without Internet

All 8 sections of this guide — complete and fully functional
Client-side search — searches all content locally
Print-to-PDF on any section
Bunker Bot AI assistant (if Ollama is running locally)
All SVG diagrams, star charts, and compass rose

Starting Ollama at Boot (No GUI Needed)

# Linux/macOS — add to crontab or startup script:
@reboot OLLAMA_MODELS=/path/to/models ollama serve &

# macOS LaunchAgent — create plist in ~/Library/LaunchAgents/
# Windows — set Ollama to start automatically via Task Scheduler

7. Test Bunker Bot in This Guide

Once Ollama is running with a model pulled, click the 🤖 Bunker Bot button in the top right corner of any page to open the chat panel. The status indicator will show:

● Checking — connecting to Ollama
● Online — Ollama running, model ready
● Offline — Ollama not reachable

Try asking Bunker Bot:

"What's the fastest way to purify water without equipment?"
"I have a deep cut that won't stop bleeding. What do I do?"
"How do I size a solar system for basic lighting and phone charging?"
"What are the signs of a failing appendix?"

💡

Bunker Bot's Purpose and Limitations

Bunker Bot is a survival-focused AI assistant — it's prompted to only answer questions related to preparedness, medical, food/water, energy, shelter, navigation, security, and community resilience. It will decline off-topic requests. All responses are generated by a local AI model and may contain errors — always cross-reference critical medical and technical information with the guide's written content and other reliable sources.

← Home