What it is

Qwen Code is an open-source AI coding agent that runs in the terminal, with additional interfaces for editors, desktop, browser, and chat. It is written in TypeScript and released under the Apache-2.0 license, and it lives in the AI and machine learning ecosystem as a development platform for agentic coding. The project is maintained by QwenLM and published on npm as @qwen-code/qwen-code, with a documentation site hosted at qwenlm.github.io. It ships as a command-line tool named qwen that opens an interactive session inside a project directory.

The concrete problem it solves is the gap between a general-purpose language model and the working context of a real codebase. Instead of copying files into a chat window, a developer starts the agent in the repository and asks it to explain the project, locate entry points, or act on the code directly. It also removes provider lock-in: the agent supports OpenAI, Anthropic, Gemini, and Qwen APIs, and can point at a third-party provider or a local model served by Ollama or vLLM, with the ability to switch at runtime. Because both the framework and the Qwen models are open-source, the tooling and the models can evolve together without a vendor controlling either side.

Key capabilities

• Agentic behavior available out of the box, including Auto-Memory, Auto-Skills, SubAgents, Agent Teams, and MCP support, with dynamic workflows and no setup step.
• Multi-protocol model support across OpenAI, Anthropic, Gemini, and Qwen APIs, plus any third-party provider or local model through Ollama or vLLM.
• Runtime switching between providers, so a session is not bound to a single API key or endpoint.
• Interfaces beyond the terminal: IDE plugins for VS Code, Zed, and JetBrains, a desktop application, a Web UI, SDKs, and chat integrations.
• Chat channel integrations for Telegram, DingTalk, WeChat, and Feishu.
• A Web UI launched through qwen serve --open, marked experimental in the documentation.
• Self-hosting of the project's own development loop, where the agent files issues, submits pull requests, reviews code, and runs tests.

Who uses it and how

• Developers working in a terminal who run qwen inside a project directory and ask the agent to explain the repository and identify where to start.
• Teams that need to route requests to a specific model provider, configuring the provider and API key inside the session with the /auth command.
• Users who prefer an editor workflow, installing the VS Code, Zed, or JetBrains integration rather than working from the command line.
• Users who want a graphical or remote surface, downloading the desktop build for macOS, Windows, or Linux, or opening the Web UI in a browser.
• Groups that coordinate work through chat platforms, connecting the agent to Telegram, DingTalk, WeChat, or Feishu.

Getting started

Installation is available through a standalone shell script for Linux and macOS, a PowerShell script for Windows, the npm package @qwen-code/qwen-code with Node.js 22 or later, or Homebrew on macOS and Linux. After installation, a developer opens a terminal in the project, runs qwen, and configures a provider and API key with /auth.

When to use it — and when not to

Choose Qwen Code when an open-source agent that can target multiple model providers, including local models, matters more than a single hosted service. A self-hoster must supply their own provider credentials, and the project carries a substantial open issue count of 1422 alongside a repository history of roughly one year, so stability expectations should match an actively iterating tool. The Web UI is documented as experimental, which makes the terminal and editor integrations the safer surfaces for daily work.

curl -fsSL https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen-standalone.sh | bash

irm https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen-standalone.ps1 | iex

npm install -g @qwen-code/qwen-code@latest

brew install qwen-code

cd /path/to/your-project
qwen

Explain this repository and show me where to start.

import asyncio

from qwen_code_sdk import is_sdk_result_message, query


async def main() -> None:
    result = query(
        "Summarize the repository layout.",
        {
            "cwd": "/path/to/project",
            "path_to_qwen_executable": "qwen",
        },
    )

    async for message in result:
        if is_sdk_result_message(message):
            print(message["result"])


asyncio.run(main())