Lesson 2: Setting up Gemma 4 with Ollama and Open WebUI on Mac

Introduction

In this lesson you'll set up a working local stack:

• Ollama as the model runtime
• Gemma 4 as the primary model
• Open WebUI as the chat interface for the whole team

1. Installing Ollama

brew install ollama
brew services start ollama
curl http://127.0.0.1:11434/api/tags

If the endpoint returns JSON, the runtime is ready.

2. Pulling the Gemma 4 Model

ollama pull gemma4
ollama run gemma4

For machines with lower RAM, prefer a quantized variant to avoid swap pressure.

3. Running Open WebUI

docker run -d \
  --name open-webui \
  -p 3000:8080 \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  -v open-webui:/app/backend/data \
  --restart unless-stopped \
  ghcr.io/open-webui/open-webui:main

Navigate to http://localhost:3000 and create the first admin account.

4. Standardizing Model Presets

Create presets by use case:

• Coding: low temperature, moderately high context
• Summarization: medium temperature, short format
• Extraction: low temperature, fixed JSON output

Keep presets in internal documentation so new team members use the correct defaults.

5. Monitoring Resources on Mac

Track 3 things:

• Memory pressure
• Swap usage
• Actual tokens/second

When swap spikes, reduce num_ctx or model size before attempting deeper optimization.

6. Quick Troubleshooting

1. Docker can't reach Ollama: use host.docker.internal.
2. Model gradually slows down: check swap and background apps.
3. UI doesn't show the model: check ollama list and OLLAMA_BASE_URL.

Exercises

• Install the full stack and capture a diagram of endpoints.
• Create 3 model presets for 3 different use cases.
• Compare speed with the same long prompt run twice in succession.

Demo Code

After installation, verify the health check endpoint:

Swagger UI auto-generates API documentation:

Source code: xdev-asia-labs/gemma-4-local-ai-engineering-on-mac

Summary