Ollama with Qwen3.5 in VS Code: setup and coding tests

This tutorial connects Ollama to Visual Studio Code through GitHub Copilot for use as a local coding assistant. Although the title says it was tested on Windows, the walkthrough uses a MacBook Pro M3 Max and includes separate advice for Windows GPUs.

The setup covers installing Ollama, adjusting its context length, downloading a model, and signing into GitHub inside VS Code. The speaker recommends at least a 64,000-token context for programming; this is his guidance rather than a universal requirement. He demonstrates Qwen3.5 9B with a reported 6.6 GB download. Qwen3.6 is discussed, but it is not the model used for the coding tests.

In Copilot's model settings, the presenter adds Ollama, keeps the local URL, and selects the downloaded model. He says a free Copilot account supports this connection, though his demonstration uses a paid account. The first request takes longer, which he attributes to VS Code's large initial context.

The tests ask the model to create hello.py, then a Python SQLite CRUD class and a separate test file. The generated files appear, and the assistant attempts to run checks; the transcript does not establish that every test passes. The speaker reports about 12 GB of RAM use and recommends a dense 9B model for Windows GPUs with 12 or 16 GB of VRAM. He also discusses oMLX caching on Mac.