Ollama Local Models#

Ollama lets you run AI models on your own computer—fully offline, fully private.

What is Ollama#

Ollama is an open-source tool for running large language models locally.

Benefits:

• Completely offline, no internet needed
• Data never leaves your computer
• Free to use, no API key needed
• Supports many open-source models

Limitations:

• Requires decent hardware
• Models generally less capable than cloud ones
• First use requires downloading models

Installing Ollama#

macOS#

# Using Homebrew
brew install ollama

# Or download installer from
# https://ollama.ai/download

Windows#

1. Go to ollama.ai/download
2. Download Windows installer
3. Run installation

Verify Installation#

ollama --version

After installation, Ollama runs as a background service.

Downloading Models#

From Terminal#

# Download Llama 3
ollama pull llama3

# Download Mistral
ollama pull mistral

# Download Qwen (good for Chinese)
ollama pull qwen2

From MoryFlow#

1. Open Settings → Model Management
2. Find the Ollama section
3. Browse available models
4. Click download

Recommended Models#

General Use#

Lower-end Hardware#

Powerful Hardware#

Configuring in MoryFlow#

1. Configure Connection#

1. Open Settings → Model Management
2. Find Ollama section
3. Confirm endpoint address (default http://localhost:11434)

2. Select Model#

Once configured, downloaded Ollama models appear in the model selector on the chat panel.

3. Start Using#

Select an Ollama model and start chatting. All processing happens locally.

Hardware Requirements#

Minimum#

• 8GB RAM
• Use 7B or smaller models

Recommended#

• 16GB RAM
• Use 7B-13B models

High Performance#

• 32GB+ RAM
• Apple Silicon (M1/M2/M3) or NVIDIA GPU
• Can run 70B models

Performance Tips#

macOS (Apple Silicon)#

Apple Silicon works great with Ollama—highly recommended.

Windows (NVIDIA GPU)#

Install the latest NVIDIA drivers and Ollama will automatically use GPU acceleration.

Not Enough RAM#

Try:

1. Switch to a smaller model
2. Close other programs
3. Set OLLAMA_NUM_PARALLEL=1 to limit concurrency

Common Issues#

Ollama Service Not Running#

# Start service
ollama serve

Model Download Failed#

1. Check network connection
2. Try using a proxy
3. Manually download model files

Response Is Slow#

1. Switch to smaller model
2. Check if GPU acceleration is working
3. Close resource-heavy programs

Chinese Characters Display Wrong#

Use Qwen2 model for better Chinese support.

Local vs Cloud#

Recommendation: Use Ollama for privacy-sensitive content, cloud models for complex tasks.