Today I learned how to run the GPT-OSS-20B model locally on my Mac using Ollama, and integrate it with VS Code as my Agent mode's default model.

The Setup

• Install Ollama on your Mac
• Pull the model: ollama pull gpt-oss-20b
• Open VS Code, navigate to the Agent mode sidebar, and click the model name at the bottom.

• Under Manage Language Models, select "Ollama"

• Select "gpt-oss-20b" from the dropdown.

Why This Matters

• Full Control: You own it end to end
• Privacy: All inference happen locally on your machine
• Cost-Free: No API costs or subscription fees
• Flexibility: Can work offline and switch between different models as needed/released

Trade-offs

The main trade-off is speed - it's notably slower than cloud-based alternatives. However, this is expected to improve as:

• Local models become more optimized
• Hardware capabilities increase
• Model architectures evolve