Qwen Code
A multi-provider AI coding agent that runs in your terminal.
Per-dimension scores and reasoning
Evidence shows permission controls (e.g., GitHub Actions least privilege, sandbox mode, user confirmation mechanisms), but static review cannot verify execution details, and no complete data flow transparency documentation is provided. Deductions: data flow transparency only partially described, sensitive data handling not detailed.
Project has extensive test scripts and CI configuration, indicating investment in consistency and dependency availability, but static review cannot verify actual runtime results. Deductions: specific failure message content not evident in provided files.
README details multiple usage scenarios (interactive, headless, IDE, desktop, SDK, IM bots) and lists capability boundaries (e.g., multi-protocol support), but not all trigger conditions are explicit. Deductions: environment fit details (e.g., specific system requirements) not fully covered.
Information architecture is clear, install notes are detailed, naming is stable (version numbers explicit), examples and FAQ links provided, license is clear, versioning has changelog scripts, maintenance responsibility has contribution guidelines. Deductions: known limitations not explicitly listed in provided files.
Output usability is high (multiple modes), marginal value is evident (comparison with Claude Code), cost-benefit is reasonable (open-source free). Deductions: no actual runtime data to support.
Claims in README (e.g., feature comparison) have external links, but static review cannot verify all claims, and facts vs. inferences not clearly separated. Deductions: cross-source corroboration insufficient.
- Static review cannot verify actual runtime behavior; dynamic testing is recommended.
- Data flow transparency documentation is incomplete; further review needed.
- Sensitive data handling details are not explicit; attention required.
What does this agent do, and when should you use it?
Qwen Code is an open-source AI coding agent with an interactive terminal UI and a headless automation mode. It supports OpenAI, Anthropic, Gemini, and Qwen APIs, plus third-party providers and local models through Ollama or vLLM, with runtime switching. Delivery options include VS Code, Zed, and JetBrains integrations, desktop applications for macOS, Windows, and Linux, and channels for Telegram, DingTalk, WeChat, and Feishu. Its experimental `qwen serve` mode exposes one shared agent session to multiple clients over HTTP+SSE using ACP. TypeScript, Python, and Java SDKs make it usable from developer tooling and automated workflows.
Run qwen to open an interactive terminal UI with rich rendering, @file references, and slash commands, then use /auth to configure a provider and API key. Run qwen -p "..." to submit a prompt without a UI for scripts, CI/CD, or batch processing. Qwen Code connects to OpenAI, Anthropic, Gemini, Qwen, third-party, or local model providers and offers MCP, SubAgents, Agent Teams, Auto-Memory, Auto-Skills, Hooks, Plan Mode, LSP integration, sandboxing, and Git worktrees. qwen serve shares an agent session with multiple clients over HTTP+SSE using ACP, while qwen channel connects Telegram, DingTalk, WeChat, or Feishu. In the Python SDK, query() accepts a prompt plus cwd and path_to_qwen_executable, and returns messages that can be consumed asynchronously.
- A software engineer who wants an interactive terminal agent with file references and slash commands while working in a repository.
- A development team invoking
qwen -pfrom CI/CD jobs, scripts, or batch-processing workflows. - A team that needs to switch at runtime among OpenAI, Anthropic, Gemini, Qwen, third-party, or local Ollama/vLLM model providers.
- A tooling developer integrating coding-agent calls into a Python, TypeScript, or Java application.
- An internal development environment that needs several clients to share one agent session over HTTP+SSE.
- A developer who wants to use a coding agent through VS Code, Zed, JetBrains, or supported messaging channels.
What are this agent's strengths and limitations?
- Supports OpenAI, Anthropic, Gemini, and Qwen APIs, plus third-party and Ollama/vLLM local models, with runtime provider switching.
- Covers interactive and headless terminal use, IDE plugins, desktop apps, an HTTP+SSE service, messaging channels, and three SDKs.
- Includes development-workflow features such as MCP, SubAgents, Agent Teams, Auto-Memory, Auto-Skills, Hooks, Plan Mode, LSP integration, sandboxing, and Git worktrees.
- The npm installation path requires Node.js 22+, and first use requires provider configuration and an API key.
- The multi-client shared-session mode exposed by
qwen serveis explicitly experimental. - The supplied material does not specify provider pricing, permission controls, default sandbox policy, or failure-recovery behavior.
How do you install or deploy this agent?
On Linux/macOS: curl -fsSL https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen-standalone.sh | bash. On Windows PowerShell: irm https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen-standalone.ps1 | iex. Alternatively, install with npm using npm install -g @qwen-code/qwen-code@latest (Node.js 22+ required), or on macOS/Linux with brew install qwen-code. Restart the terminal after installation, then configure a provider and API key with /auth on first launch.
How do you use this agent?
Run qwen to launch the interactive terminal UI, then run /auth to configure a provider and API key. For headless execution, use qwen -p "Summarize the repository layout.". Start the experimental shared-session service with qwen serve, or connect messaging channels with qwen channel. From Python, call query("Summarize the repository layout.", {"cwd":"/path/to/project","path_to_qwen_executable":"qwen"}) and asynchronously iterate over the returned messages for the result.
How does this agent compare with similar options?
Qwen Code names Claude Code as a comparison point and states corresponding support for SubAgents, Agent Teams, MCP, Plan Mode, LSP integration, sandboxing, Git worktrees, IDE plugins, SDKs, headless mode, and session management. Its listed differentiators are open-source models and framework, multi-protocol provider support, Agent Arena, the qwen serve shared-agent service, and Telegram, DingTalk, WeChat, and Feishu channels.
FAQ
Do I have to use Qwen models?
What is required for first use?
qwen, then configure a model provider and API key with /auth. The npm installation route also requires Node.js 22+.Can it run in automation or CI/CD?
qwen -p "..." is the headless mode, listed for scripts, CI/CD, and batch processing.Can my own application call it?
query() and consumes an asynchronous message stream.