How to Run Local AI Offline on Your Computer: Complete Ollama & Google Gemma 2 Setup Guide

Imagine being on an airplane with no Wi-Fi, in a secure office building with strict data confidentiality policies, or at home with no internet connection—and still having a full artificial intelligence assistant ready to write code, polish resumes, summarize documents, and generate SQL queries.

With Ollama and modern open-weights models like Google Gemma 2 2B, running AI 100% locally and offline on your computer has never been easier.

In this beginner-friendly tutorial, we walk you step-by-step through installing Ollama on macOS, Windows, or Linux, downloading Google Gemma 2, and connecting it seamlessly to your browser with zero subscription fees.

Step 1: Download & Install Ollama

Ollama is a free, open-source utility that manages local AI models.

1. Head to the official Ollama download page: ollama.com/download.
2. Choose your operating system:
   • macOS: Download the .zip, unzip it, drag Ollama to your Applications folder, and launch it.
   • Windows: Download the installer (OllamaSetup.exe), run the setup wizard, and follow the prompts.
   • Linux: Run the one-line terminal installer:

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

Once installed, Ollama runs quietly in your system tray or background service.

Step 2: Download Google Gemma 2 2B (Lightweight & Fast)

Google released Gemma 2—an open-weights family of models derived from the same research behind Gemini. The 2B (2-Billion parameter) model is exceptional because it delivers smart responses while occupying only ~1.6 GB of disk space and running smoothly on laptops with just 4GB to 8GB of RAM.

Open your Terminal (macOS/Linux) or PowerShell (Windows) and type:

ollama run gemma2:2b

Ollama will automatically download the model files and launch an interactive chat prompt directly inside your terminal. You can type hello to verify it works! Press Ctrl + D or type /bye to exit the terminal session.

Want an even lighter model? You can also run Meta’s ultralight model:

ollama run llama3.2:1b

This model consumes just 1.3 GB of disk space and runs with blazing speed.

Step 3: Start Ollama with Browser CORS Permissions

Web browsers use security policies called CORS (Cross-Origin Resource Sharing). By default, browsers prevent web applications like Everyday Used Apps from sending requests to http://localhost:11434 unless Ollama explicitly allows it.

To allow your browser to communicate with Ollama, you must launch Ollama with the OLLAMA_ORIGINS environment variable set:

On macOS & Linux (Terminal):

OLLAMA_ORIGINS="*" ollama serve

On Windows (PowerShell):

$env:OLLAMA_ORIGINS="*"
ollama serve

Setting It Permanently (Optional):

If you don’t want to type this command every time:

• macOS: Add export OLLAMA_ORIGINS="*" to your ~/.zshrc or ~/.bash_profile.
• Windows: Search for “Environment Variables” in Start Menu > System Properties > Add User Variable OLLAMA_ORIGINS with value *.

Step 4: Connect Ollama to Everyday Used Apps

Now that Ollama is serving on port 11434, connecting it to our web tools takes just two clicks:

1. Open any AI tool on Everyday Used Apps (e.g., AI Code Assistant or AI Document Q&A).
2. Click the “Enable AI” pill button in the top navigation header.
3. Switch to the Google Gemma 2 & Local tab in the vault dialog.
4. Verify the Local Endpoint is set to http://localhost:11434.
5. Select Google Gemma 2 2B (Safe • Low RAM) from the model dropdown.
6. Click Test Connection to confirm latency (usually 5ms to 15ms).
7. Click Save & Enable AI.

That’s it! Your browser tools are now powered entirely by the AI running on your computer.

Step 5: Test Air-Gapped Offline Operation

To verify that your AI is truly offline and private:

1. Disconnect your computer from Wi-Fi or unplug your Ethernet cable.
2. Go back to your browser tab with Everyday Used Apps.
3. Paste a block of Python code, SQL query, or email draft into any AI tool.
4. Click Generate or Analyze.
5. Watch the response stream back in real-time—with zero internet connection!

Preventing Laptops from Freezing: The 3 Rules

Running local AI is empowering, but follow these 3 best practices to maintain a smooth system:

1. Don’t Run Models Larger than Your RAM:
   • 8GB RAM Laptop: Stick to Gemma 2 2B or Llama 3.2 1B.
   • 16GB RAM PC: You can safely run Gemma 2 9B or Llama 3.1 8B.
   • Avoid 70B models locally unless you possess multiple dedicated high-end GPUs.
2. Close Heavy Applications: Before running complex local reasoning, close video editors, high-memory IDEs, and 50+ background browser tabs.
3. Use Built-In Watchdogs: Everyday Used Apps includes a 45-second auto-watchdog that cancels tasks if your hardware hangs, ensuring your browser remains responsive.

Enjoy total data sovereignty, zero subscription bills, and private artificial intelligence!