Qwen Code

Overview#

Qwen Code is a command-line AI agent tool developed by the QwenLM team, built with TypeScript and optimized specifically for the Qwen3-Coder model. It can understand large codebases, execute shell commands, operate the file system, and connect to external data sources via MCP (Model Context Protocol).

Core Positioning: Terminal-first AI coding agent with Prompt adaptation and parsing optimization for Qwen3-Coder models.

Core Capabilities#

Code Development Assistance#

• Code Generation & Editing: Generate code snippets, functions, or complete files from natural language descriptions
• Codebase Navigation: Answer questions about project structure and module dependencies through semantic analysis
• Debugging & Fixing: Parse error stacks, automatically locate issues, and provide fix suggestions
• Test Generation: Automatically generate unit tests for specified modules
• Documentation Maintenance: Assist in generating README or code comments

Agent & Automation#

• Agentic Workflow: Support defining Skills and SubAgents for multi-step complex tasks
• Shell Interaction: Execute terminal commands under permission control (requires user confirmation)
• Automation Scripts: Batch process lint fixes, code formatting, and other repetitive tasks

Connection & Extension#

• MCP Protocol: Connect as a client to external tools like Google Drive, Figma, Slack
• Multi-Model Support: Compatible with OpenAI, Anthropic, Gemini API formats
• SDK Support: Provides TypeScript SDK and Java SDK

Key Features#

• Multi-Protocol Support + Free Quota: Supports OpenAI / Anthropic / Gemini compatible APIs; Qwen OAuth provides 1,000 free requests daily
• Open Source Co-evolution: Both framework and Qwen3-Coder model are open source, iterating synchronously
• Terminal-First + IDE Friendly: Built for command-line users, optional VS Code, Zed, JetBrains IDE integration
• Multi-Language UI: zh-CN, en-US, ru-RU, de-DE, etc.

Typical Use Cases#

• Build features from natural language descriptions
• Debug and fix issues (describe bugs or paste error messages)
• Navigate and understand any codebase
• Automate repetitive tasks (fix lint, resolve merge conflicts, write release notes)
• Interactive Git operations (view changes, commit, create branches, resolve merge conflicts)
• Code refactoring and test generation

Installation#

Prerequisites: Node.js ≥ 20

# NPM (Recommended)
npm install -g @qwen-code/qwen-code@latest

# Homebrew (macOS, Linux)
brew install qwen-code

# Quick Install Script (Linux / macOS)
curl -fsSL https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen.sh | bash

Quick Start#

# Start interactive session
qwen

# In-session commands
/help      # View help
/auth      # Configure authentication (Qwen OAuth recommended, 1000 free daily)
/model     # Switch model
/mcp       # Manage MCP servers
/tools     # Show available tools

Architecture#

Adopts Frontend (CLI) + Backend (Core) separation architecture:

• packages/cli: Input handling, history management, display rendering, theme/UI customization, config management
• packages/core: API client, Prompt building, tool registration and execution, state management
• packages/sdk-typescript: TypeScript SDK for programmatic access
• packages/sdk-java: Java SDK
• packages/vscode-ide-companion: VS Code extension
• packages/zed-extension: Zed editor extension

Performance Benchmarks#

Terminal-Bench results:

• Qwen Code + Qwen3-Coder-480A35: 37.5%
• Qwen Code + Qwen3-Coder-30BA3B: 31.3%