Qwen Code
Run the open-source Qwen Code agent against any OpenAI-compatible model inside a Helix sandbox.
Qwen Code is an open-source coding agent built around the Qwen models, but its harness is provider-agnostic: it talks OpenAI-compatible JSON, so any model served behind that API (including non-Qwen ones) works. Helix runs the qwen CLI as a custom ACP agent inside the sandbox desktop, surfaced through Zed's agent panel.
Pick Qwen Code when you want a fully open-source agent stack, you're running local models on a Helix runner, or you need to point the agent at an OpenAI-compatible endpoint that isn't Anthropic or OpenAI proper.
Selecting Qwen Code for a project
Set the runtime to qwen_code and choose a model:
apiVersion: helix.ml/v1alpha1
kind: Project
metadata:
name: my-app
spec:
description: "App using Qwen Code"
repositories:
- url: "https://github.com/org/my-app"
branch: main
primary: true
agent:
name: "Qwen Coder"
runtime: qwen_code
model: qwen3-coder-480b
provider: helixHow the model is wired up
Helix routes the agent through its own /v1 proxy in OpenAI-compatible mode, prefixing the model name with the provider you chose. That gives you one consistent path regardless of where the model actually lives:
- Local models — point
providerat a Helix runner running Qwen, Llama, or any other supported model. Traffic never leaves your infrastructure. - Hosted OpenAI-compatible providers — Together, OpenRouter, Groq, vLLM, anywhere that speaks the standard.
- OpenAI proper — works the same way; just set
provider: openai.
Configure providers under Account → AI Providers.
When to pick Qwen Code
- You want a fully open-source agent harness, not a vendor CLI.
- You want to run the agent against local models for data-privacy or air-gapped reasons.
- You're already invested in OpenAI-compatible tooling and want the agent to match.
See Claude Code if you want Anthropic's official agent with Claude models, Goose for project-scoped recipes, or Zed Agent for the built-in in-editor agent.