3. Configuration (settings.json)

The qwen code is extremely strict about the JSON schema. If it's incorrect, the app will rename the file to settings.json.corrupted.

Create/edit the file at:

• Windows: %USERPROFILE%\.qwen\settings.json
• Linux/macOS: ~/.qwen/settings.json

Use this exact structure:

4. Launcher Scripts

Since the CLI sometimes ignores file-based environment variables, use a launcher script for a stable connection.

• Windows (start_qwen.ps1)

• Linux/macOS (start_qwen.sh)

https://qwenlm.github.io/qwen-code-docs/en/users/configuration/model-providers/ This link is provided by qwen code itself. There is outdated information there. Everything written on the website is incorrect, I checked it and it doesn't work.

5. Pro Tips for smooth execution

• Reasoning over Looping: If the model feels "insecure" or keeps asking questions, switch to an RL-tuned model. As tested, a 9B RL model can handle complex file analysis that standard models fail to process. ✦ qwen3.5-9b-sushi-coder-rl is preferred LLM model for plugins TS
• Tool Usage: If the model analyzes files but doesn't apply changes, tell it: "Analyze the directory and rewrite with a clean, typed implementation now."
• Temperature: Set Temperature to 0.0 or 0.1 in LM Studio for more predictable, structured code output.

PS: But for VS Code, a different settings.json worked.

The file settings.json in utf-8.

🔌 Qwen Code Companion is the pugin for VS Code.

This is a VS Code plugin developed by Alibaba Group. It's an intelligent assistant based on the Large Language Model (LLM) that helps you write and debug code right in the editor. Even more settings https://github.com/QwenLM/qwen-code also see authentication guide