How to connect Qwen to VS Code: Companion, Cline and Continue

There are two ways to connect Qwen to VS Code, and they solve slightly different problems. The first is the official Qwen Code Companion extension, which brings the Qwen Code agent into the editor. The second is third-party extensions such as Cline, Kilo Code or Continue, which only need an address, a key and a model name. Below are both options set up with one Zukko key, with prices as of October 18, 2026.

Which path to choose

If you already use Qwen Code in the terminal and want edits highlighted in the editor, take Companion. If you just want an agent chat in the editor without installing another program, take Cline, Kilo Code or Continue.

Path 1: Qwen Code Companion

According to the project docs, the extension (in Beta) provides a dedicated Qwen Code panel, @ mentions of files, attaching files and images, conversation history, several parallel sessions and an auto-accept mode for edits. It requires VS Code 1.96.0 or newer. It installs from the Marketplace under the identifier qwenlm.qwen-code-vscode-ide-companion (search for "Qwen Code Companion").

Steps:

1. Install Qwen Code itself: npm install -g @qwen-code/qwen-code@latest (Node.js 22+ required).
2. Connect a model. The shortest way for a Zukko key:

export OPENAI_API_KEY="fetch_your_key"
export OPENAI_BASE_URL="https://api.zukko.pro/v1"
export OPENAI_MODEL="qwen3.8-max"

1. Install the extension from the Marketplace. The docs also mention the /ide install command inside Qwen Code, and the panel can be launched from the command palette ("Qwen Code: Run").
2. Open the project folder. The docs advise keeping a single workspace folder in the window.

If Companion does not connect, the Qwen Code troubleshooting section says: open exactly one folder, restart the integrated terminal after installing the extension (so it receives the QWEN_CODE_IDE_WORKSPACE_PATH and QWEN_CODE_IDE_SERVER_PORT variables), and in a container check that host.docker.internal resolves.

What we could not confirm from the docs: whether the side panel reads the same environment variables as the terminal or only the ~/.qwen/settings.json file. The safest way is to describe the model in settings.json, so the setup is shared by the terminal and the panel. A persistent config is covered in Qwen Code CLI: connecting↗, and if something goes wrong, check the VS Code integration section in the Qwen Code documentation.

Path 2: Cline, Kilo Code or Continue

These extensions run their own agent and do not need Qwen Code. Roo Code, often recommended in this group before, was shut down on May 15, 2026 according to its developers, so it is not listed here. The setup scheme is the same everywhere: provider, address, key, model. Per-extension details are in the Zukko documentation↗.

A Continue example:

models:
  - name: Qwen 3.8 Max
    provider: openai
    model: qwen3.8-max
    apiKey: fetch_your_key
    apiBase: https://api.zukko.pro/v1

More details: Cline↗, Continue↗. Agent extensions need a model that calls tools reliably; if the chosen model behaves oddly (describes actions instead of performing them), switch it and retry.

Which Qwen model to take

Two Qwen models are available through Zukko. Prices per 1M tokens as of October 18, 2026:

If you need to attach screenshots, take qwen3.8-max: qwen3.8-flash has no image input. Other catalog models connect exactly the same way; only the id changes.

An example edit on a small project: 300K input tokens and 30K output. Formula: input × input price + output × output price.

• qwen3.8-max: 0.3 × 0.088 + 0.03 × 0.238 = 0.0264 + 0.00714 = $0.03354.
• qwen3.8-flash: 0.3 × 0.038 + 0.03 × 0.075 = 0.0114 + 0.00225 = $0.01365.
• claude-sonnet-5-5 for comparison: 0.3 × 0.4 + 0.03 × 2 = 0.12 + 0.06 = $0.18.

Agent extensions resend the whole history at each step, so real input is larger. Watch the spend in the "Usage" section of the dashboard. We cannot say in advance which model handles your code better: try both on the same task.

Checking the key outside the editor

If an extension is silent, test the key with a command:

curl https://api.zukko.pro/v1/chat/completions \
  -H "Authorization: Bearer fetch_your_key" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-flash","messages":[{"role":"user","content":"ping"}]}'

If curl answers and the editor does not, look for the error in the extension settings (address without /v1, a stray space in the key, a wrong model name). If curl does not answer either, the error code tells the cause (401 wrong key, 402 insufficient balance, 404 unknown model).

Create the key in the dashboard↗; in the key settings you can set a daily dollar limit so the editor agent does not overspend.

FAQ

Does VS Code have built-in Qwen support?

No, you need an extension. The official one is called Qwen Code Companion and works together with Qwen Code. Alternatives: Cline, Kilo Code, Continue.

Is Companion free?

The extension and Qwen Code itself are open, but you pay for the model: through a provider or gateway key. The free Qwen OAuth login, according to the docs, was discontinued on April 15, 2026.

Can these extensions use Claude or GPT instead of Qwen?

Yes. They all accept any id from the catalog: claude-sonnet-5-5, gpt-6.1-sol and others. Note that in the OpenAI format Claude input is billed at the full input price, with no cache discount.

How do I pay from Russia?

The dollar balance is topped up via SBP (the payment page adds 8.5% on top), MasterCard or cryptocurrency, with a minimum top-up of $1.