3.6 Local Free AI: Ollama

Ollama is a free tool that supports running AI models locally. Using Ollama, you can use AI completely free and offline without purchasing an API Key—or subscribe to an Ollama plan for greater model compute capacity with high or unlimited request limits.

Core Configuration Information

When configuring Ollama, you need to find the following core information:

• Host IP Address: Usually starts with 192.168 (e.g., 192.168.1.2)
• Base URL Format: http://192.168.1.2:11434/api

Important Note

• Must use http protocol (not https)
• Must use port 11434
• Must add /api at the end

Example: http://192.168.1.2:11434/api (where 192.168.1.2 is your computer's IP)

---

Steps (macOS Example)

Step 1: Download and Install Ollama

Visit https://ollama.com to download and install Ollama

Step 2: Download Model

Open terminal and download a small model for testing (e.g., deepseek-r1:1.5b, about 1GB):

ollama pull deepseek-r1:1.5b

Step 3: Run Model

After download completes, run the model:

ollama serve
ollama run deepseek-r1:1.5b

First startup will be slower, afterwards you can chat with the model directly in terminal for testing

Step 4: Enable LAN Access

Enable LAN access in Ollama settings to allow phone to connect to Ollama on your computer

Why is this needed?

• MiniTavern runs on phone and needs to access Ollama on computer via LAN
• By default, Ollama only allows local access
• After enabling, devices on the same Wi-Fi can access it

If you prefer the command line and did not install the Ollama GUI app, you can skip the client setting in this step and follow Enable LAN access via CLI (0.0.0.0) below.

Step 5: Get Computer IP Address

Check your computer's IP address in system settings (usually starts with 192.168)

macOS IP Check Method:

System Settings → Network → Wi-Fi → Details → TCP/IP → IPv4 Address

Or use terminal command:

ifconfig | grep "inet " | grep -v 127.0.0.1

Step 6: Configure in MiniTavern

6.1 Select Other LLM Provider

Open MiniTavern → Settings → LLM Settings → AI Provider → Select Other

6.2 Enter Ollama Base URL

Enter Ollama Base URL: http://your-computer-IP:11434/api

Example: http://192.168.1.2:11434/api

Important Note

• Must use http (not https)
• Port must be 11434
• Must add /api at the end

Click Get Model List

6.3 Test Connection

After selecting model, click Test Connection to verify configuration is correct

If unable to connect:

• Check if Ollama is running
• Confirm IP address is correct
• Ensure phone and computer are on same Wi-Fi
• Check if Base URL format is correct

6.4 Add Complete

After successful test, click save, model addition complete

---

Install Ollama Using Brew

brew install ollama

Enable LAN access via CLI (0.0.0.0)

This section uses macOS as an example and shows how to change Ollama’s listen address to 0.0.0.0 from the command line—equivalent to Expose to the Network in the client.

1. If you did not install the Ollama app and only installed the CLI with brew install ollama, the default start only shares model access on this computer, listening on 127.0.0.1. That address cannot be reached by other devices on the LAN.
2. Stop the current service with pkill ollama, then restart serve bound to 0.0.0.0.
3. If you previously downloaded and auto-started the Ollama app and it is still running in the background, fully quit the app first; otherwise you will see that the port is already in use.
4. Then run:

OLLAMA_HOST=0.0.0.0:11434 ollama serve

This starts a local serve that is reachable on the LAN. In MiniTavern, keep using your computer’s LAN IP for the Base URL, for example: http://192.168.1.2:11434/api.

---

Common Ollama Commands

# View all downloaded models
ollama list

# Download tutorial example model
ollama pull deepseek-r1:1.5b

# Start Ollama service (localhost 127.0.0.1 only)
ollama serve

# Start Ollama service and expose to LAN (0.0.0.0)
OLLAMA_HOST=0.0.0.0:11434 ollama serve

# Run specified model in new process
ollama run deepseek-r1:1.5b

# Stop Ollama service
pkill ollama
# Or press Ctrl + C if running in the foreground

---

FAQ

Q: Poor conversation quality?

A:

• Try using larger models (7B or above)
• Try different character cards

---

Related Resources

• Ollama Official Documentation
• Ollama Model Library