qwen-code

Qwen Code Example

Run Qwen Code inside an OpenSandbox container through an OpenAI-compatible endpoint.

Start OpenSandbox server [local]

Pre-pull the code-interpreter image (includes Node.js):

docker pull sandbox-registry.cn-zhangjiakou.cr.aliyuncs.com/opensandbox/code-interpreter:v1.0.2

# use docker hub
# docker pull opensandbox/code-interpreter:v1.0.2

Then start the local OpenSandbox server, stdout logs will be visible in the terminal:

uv pip install opensandbox-server
opensandbox-server init-config ~/.sandbox.toml --example docker
opensandbox-server

Create and Access the Qwen Sandbox

# Install OpenSandbox package
uv pip install opensandbox

# Export provider settings
export API_KEY=your-api-key
export BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
export MODEL_NAME=qwen3-coder-plus

# Run the example
uv run python examples/qwen-code/main.py

The script installs Qwen Code (npm install -g @qwen-code/qwen-code@latest) at runtime, writes a project-local .qwen/settings.json inside the sandbox, and runs qwen -p "Compute 1+1 and reply with only the final number." in headless mode. The API key is injected only through the API_KEY environment variable and is not written into the repository.

Environment Variables

• SANDBOX_DOMAIN: Sandbox service address (default: localhost:8080)
• SANDBOX_API_KEY: API key if your server requires authentication (optional for local)
• SANDBOX_IMAGE: Sandbox image to use (default: sandbox-registry.cn-zhangjiakou.cr.aliyuncs.com/opensandbox/code-interpreter:v1.0.2)
• API_KEY: API key for the OpenAI-compatible provider used by Qwen Code (required)
• BASE_URL: OpenAI-compatible base URL (default: https://dashscope.aliyuncs.com/compatible-mode/v1)
• MODEL_NAME: Model name for Qwen Code (default: qwen3-coder-plus)

References

• Qwen Code - Official repository
• Qwen Code Authentication - Provider configuration reference