A method for local LLM development using a MacBook Pro (M5 model, 48GB RAM) utilizing Ollama, OpenCode, and a Docker sandbox has been introduced.

Ollama is characterized by its easy model management and stability, supporting mlx for high-speed operation on Apple Silicon. By combining this with OpenCode, a development framework, and executing it within a Docker sandbox (sbx), a secure development environment can be constructed that avoids unexpected impacts on the system caused by the models.

As a specific example, a configuration is shown where Qwen 3.8 27B (mxfp8 format) serves as the main working model, and Gemma 4 31B (mxfp8 format) is used as a model for inference and analysis, coordinated through the OpenCode configuration file (opencode-local.json). This enables the development of web applications leveraging powerful models running locally.

Source: