Qwen CLI - Community Fork

English | 简体中文

A community fork of Google's Gemini CLI, modified to work with Qwen models from Alibaba Cloud

This repository contains a community-maintained fork of Google's Gemini CLI, modified to work with Qwen models. It's a command-line AI workflow tool that connects to your tools, understands your code and accelerates your workflows using Qwen's powerful language models.

Attribution

This project is a community fork of Google's Gemini CLI, originally developed by Google. We've modified it to work with Qwen models from Alibaba Cloud while maintaining the excellent architecture and functionality of the original project.

Original Project: google-gemini/gemini-cli
Original License: Apache License 2.0
This Fork: Community-maintained, independent project

What You Can Do

With the Qwen CLI you can:

• Query and edit large codebases using Qwen's 131k+ token context window
• Generate new apps from PDFs or sketches, using Qwen's multimodal capabilities
• Automate operational tasks, like querying pull requests or handling complex rebases
• Use tools and MCP servers to connect new capabilities
• Advanced vision capabilities with Qwen VL models for image analysis and processing
• Switch between multiple Qwen models with an intuitive dialog interface
• Use the CLI in Chinese or English with full localization support
• Multi-agent task coordination for parallel processing and complex workflows
• Assistant Mode with a modern web interface for non-technical users
• 🎬 Media Generation with Wan Models - Create videos from text/images and transform images (Assistant Mode only)

🌐 Assistant Mode (New!)

Launch Qwen CLI with a user-friendly web interface perfect for non-technical users:

# Start in assistant mode
node bundle/qwen.js --assistant

This opens a modern chat interface in your browser with:

• 📎 File Upload Support - Upload images, videos, documents, and any file type
• 🖼️ Visual File Processing - Qwen analyzes images and documents in context
• 💬 Familiar Chat Interface - Similar to ChatGPT/Claude for ease of use
• 🔧 Full Tool Access - All CLI capabilities available through the web interface
• 📁 Session-Based Storage - Temporary file management with automatic cleanup
• 🌓 Dark Mode Support - Automatic theme detection based on system preferences

Perfect for team members who need AI assistance without command-line expertise!

🎬 Media Generation with Wan Models (Assistant Mode Exclusive)

In Assistant Mode, you gain access to Alibaba's powerful Wan media generation models:

Available Tools:

• generate_video - Create videos from text descriptions with multilingual support
• transform_image - Apply artistic transformations (cartoon, oil painting, anime, etc.)
• generate_image_to_video - Convert static images into dynamic videos
• search_wan_models - Discover available models and their capabilities

Example Use Cases:

• Create product demonstration videos from descriptions
• Transform product photos into different artistic styles
• Generate marketing content with animations
• Convert uploaded images into engaging videos

Quick Example:

# In Assistant Mode
> Transform this product photo into a cartoon style
> Create a 10-second video showing "a modern office with people collaborating"
> Convert my uploaded image into a 5-second video with a waving motion

For detailed Wan tools documentation, see the Wan Integration Guide.

Quickstart

Note: This is a community fork. Installation requires building from source as it's not published to npm registries.

1. Prerequisites: Ensure you have Node.js version 18 or higher installed.

2. Clone and Build:

   git clone https://github.com/[your-username]/qwen-cli-fork
   cd qwen-cli-fork
   npm install
   npm run build
   npm run bundle

3. Run the CLI:

   node bundle/qwen.js

4. Set up Authentication: Configure your Qwen API key and base URL from Alibaba Cloud DashScope:

   export DASHSCOPE_API_KEY="your-api-key-here"
   # or
   export QWEN_API_KEY="your-api-key-here"
   
   # Base URL (choose based on your region):
   # For Chinese mainland users:
   export QWEN_BASE_URL="https://dashscope.aliyuncs.com/api/v1"
   
   # For international users:
   export QWEN_BASE_URL="https://dashscope-intl.aliyuncs.com/api/v1"

5. Pick a color theme and start using the CLI!

You are now ready to use the Qwen CLI fork!

Getting Your Qwen API Key

1. Create an account at Alibaba Cloud DashScope
2. Navigate to the API Keys section in your dashboard
3. Create a new API key for DashScope services
4. Set it as an environment variable as shown above

Important Regional Configuration:

• Chinese mainland users: Use https://dashscope.aliyuncs.com/api/v1
• International users: Use https://dashscope-intl.aliyuncs.com/api/v1

The CLI will automatically detect and use the appropriate endpoint based on your QWEN_BASE_URL setting. If not set, it defaults to the international endpoint.

Supported Models

This fork supports the following Qwen models:

• qwen-turbo-latest - Fast model with 1M context window
• qwen3-235b-a22b - Most capable model with 131k context window
• qwen-vl-plus-latest - Vision model for image analysis (32k context)

For detailed model specifications, see our model documentation.

Key Features

🌐 Multilingual Support

Switch between English and Chinese with the /lang command:

• Full UI translation (menus, help text, command descriptions)
• System prompts automatically adapt to selected language
• Settings are persisted across sessions
• Changing language restarts the chat session so the new system prompt is used

🤖 Dynamic Model Switching

Easily switch between Qwen models with the /model command:

• Interactive dialog showing all available models
• Model specifications (context window, output tokens, vision support)
• Thinking Mode Toggle - Enable/disable chain-of-thought reasoning without restart
• Seamless switching without restarting the CLI

⌨️ Consistent User Experience

All configuration commands now use a unified dialog pattern:

• /theme - Change visual themes
• /auth - Switch authentication methods
• /lang - Change interface language
• /model - Switch Qwen models

🔍 Web Search Integration

The CLI includes powerful web search capabilities through MCP (Model Context Protocol) servers:

Quick Setup:

# In the CLI, run:
/setup-mcp websearch

This automatically configures DuckDuckGo search for privacy-focused web search. You can then use web search in your conversations:

> Search for the latest TypeScript 5.7 features and help me understand them

Manual Setup (Optional): If you prefer manual configuration, add to your .qwen/settings.json:

{
  "mcpServers": {
    "duckduckgo": {
      "command": "npx",
      "args": ["-y", "@oevortex/ddg_search"]
    }
  }
}

Management Commands:

• /mcp - Check configured MCP servers and their status
• /setup-mcp - Show available MCP setup options
• /tools - List all available tools including web search

🔧 Dynamic MCP Server Management

The CLI now features advanced MCP server management capabilities with interactive dialogs and AI-powered installation:

Interactive MCP Menu:

# Launch the interactive MCP management interface
/mcp

This opens a comprehensive menu with four options:

• 🔍 Browse Servers - Explore MCP servers by category (Development, Search, Database, etc.)
• 🔎 Search Servers - Find servers by name or functionality
• 📦 Install Server - Configure and install servers with scope and trust settings
• 📋 List Servers - View currently configured MCP servers

AI-Powered Installation: The CLI can now install MCP servers automatically when you need new capabilities:

> I need to search the web for information
# AI will automatically install and configure DuckDuckGo search server

> Help me manage my PostgreSQL database
# AI will find and install the PostgreSQL MCP server

> I want to integrate with Slack
# AI will discover and set up the Slack MCP server

Built-in MCP Server Catalog: The CLI includes a curated catalog of popular MCP servers:

• DuckDuckGo Search - Privacy-focused web search
• File Manager - Advanced file operations
• GitHub Integration - Repository and issue management
• PostgreSQL - Database operations
• Slack - Team communication
• AWS S3 - Cloud storage management
• Memory Server - Persistent context storage
• Cloud Storage - Multi-cloud file operations

MCP Tools for AI:

• add_mcp_server - Install and configure MCP servers programmatically
• search_mcp_servers - Discover available servers by category or keyword

Example Workflows:

> Find and install all development-related MCP servers
# AI will search for and present development tools

> Set up MCP servers for a web development project
# AI will install servers for search, database, and file management

> Show me what communication servers are available
# AI will list Slack, Discord, and other communication integrations

🛠️ Enhanced Shell Command Handling

Improved execution for commands with long outputs:

• Smart Output Buffering - Handles commands like npm run dev without memory exhaustion
• Automatic Timeouts - Dev server commands get appropriate timeouts (10s for dev servers, 30s default)
• Output Truncation - Large outputs automatically truncated at 1MB with clear indicators
• Real-time Updates - See command output as it streams

🚀 Multi-Agent Task Coordination

Revolutionary parallel processing capabilities that enable sophisticated workflow automation:

Core Multi-Agent Tools:

• spawn_sub_agent - Launch independent Qwen CLI instances for isolated tasks
• delegate_task - Split complex work into coordinated subtasks with parallel/sequential execution
• aggregate_results - Combine and analyze outputs from multiple agents

Key Capabilities:

• Parallel Processing: Execute up to 5 concurrent agents simultaneously
• Intelligent Scheduling: Priority-based task queue with timeout management
• Resource Management: Automatic load balancing and memory optimization
• Result Synthesis: Multiple aggregation formats (summary, merge, compare, analyze)

Example Multi-Agent Workflows:

> Analyze this entire codebase using multiple agents: one for security issues, one for performance bottlenecks, and one for code quality metrics

> Set up a new project with parallel tasks: create frontend structure, set up backend APIs, configure database, and write documentation

> Run comprehensive testing: unit tests, integration tests, and performance benchmarks all in parallel, then aggregate the results

For detailed usage examples, see the Multi-Agent System Documentation.

Examples

Once the CLI is running, you can start interacting with Qwen from your shell.

You can start a project from a new directory:

cd new-project/
node /path/to/bundle/qwen.js
> Write me a Discord bot that answers questions using a FAQ.md file I will provide

Or work with an existing project:

git clone https://github.com/[your-username]/qwen-cli-fork
cd qwen-cli-fork
node bundle/qwen.js
> Give me a summary of all of the changes that went in yesterday

Next steps

• Learn how to contribute to this community fork
• Explore the available CLI Commands
• If you encounter any issues, review the Troubleshooting guide
• For more comprehensive documentation, see the full documentation
• Take a look at some popular tasks for more inspiration
• Check out the original Gemini CLI for updates and comparisons

Troubleshooting

Head over to the troubleshooting guide if you're having issues.

Popular tasks

Explore a new codebase

Start by cding into an existing or newly-cloned repository and running the Qwen CLI.

> Describe the main pieces of this system's architecture.

> What security mechanisms are in place?

Work with your existing code

> Implement a first draft for GitHub issue #123.

> Help me migrate this codebase to the latest version of Java. Start with a plan.

Automate your workflows

Use MCP servers to integrate your local system tools with your enterprise collaboration suite.

> Make me a slide deck showing the git history from the last 7 days, grouped by feature and team member.

> Make a full-screen web app for a wall display to show our most interacted-with GitHub issues.

Multi-agent coordination

Leverage the power of parallel processing for complex tasks.

> Use multiple agents to simultaneously refactor the authentication system, update the API documentation, and create comprehensive tests

> Analyze this monorepo with dedicated agents for each microservice, then aggregate findings into a single architecture report

> Set up a complete CI/CD pipeline: one agent configures Docker, another sets up GitHub Actions, and a third creates deployment scripts

Interact with your system

> Convert all the images in this directory to png, and rename them to use dates from the exif data.

> Organise my PDF invoices by month of expenditure.

📊 Telemetry & Conversation Logging

The Qwen CLI includes a comprehensive telemetry system for analyzing interactions and optimizing prompt performance:

Conversation Logs

View your AI conversations in a human-readable format for analysis and debugging:

# View recent conversations
npm run logs

# View last 5 sessions
npm run logs -- --last 5

# Follow logs in real-time
npm run logs:tail

# Export conversations to markdown
npm run logs:export

# Search for specific content
npm run logs -- --search "function"

# View specific session
npm run logs -- --session "session-id-here"

Features

• 📝 Complete Conversation History - Captures both user prompts and AI responses
• 🔧 Tool Call Tracking - Detailed logging of function executions and results
• 🎨 Color-Coded Output - User messages, AI responses, and tool calls distinctly colored
• 📊 Session Analytics - Duration, message counts, and tool usage summaries
• 🔍 Search & Filter - Find specific conversations or content across sessions
• 📤 Export Options - Console, Markdown, and JSON output formats

Prompt Analysis

Analyze conversation patterns for prompt optimization:

# Analyze prompt patterns and generate insights
npm run telemetry:analyze

# Set up telemetry collection
npm run telemetry

Example Output

Session: abc123-def456-789
Started: 10:30:25 PM 7/3/2025
Duration: 5m 42s
Messages: 3 prompts, 8 tools (read_file, edit, write_file)

👤 User [10:30:25 PM]:
Fix the authentication bug in the login component

🤖 Qwen (qwen3-235b-a22b) [2.3s]:
I'll help you fix the authentication bug. Let me first examine the login component to understand the issue.

🔧 Tool: read_file [25ms]
Args: {"file_path": "/src/components/Login.tsx"}
Status: ✅ Success

Configuration

Telemetry is automatically enabled in your .qwen/settings.json:

{
  "telemetry": {
    "enabled": true,
    "target": "local",
    "logPrompts": true,
    "otlpEndpoint": "http://localhost:4317"
  }
}

For detailed telemetry documentation, see:

• Telemetry Usage Guide - Basic usage and configuration
• Conversation Logs Guide - Advanced log viewing and analysis
• Prompt Tuning Guide - Optimizing prompts with telemetry data

Technical Improvements

This fork includes several technical enhancements over the original:

• Complete Telemetry Rebrand - All telemetry systems properly rebranded from Gemini to Qwen
• Enhanced Authentication - Automatic retry with authentication refresh on API failures
• Improved Settings Persistence - Model and thinking mode preferences saved across sessions
• Better Memory Management - Shell commands with large outputs handled efficiently
• Dynamic Configuration - Change models and thinking modes without losing conversation context
• Comprehensive Logging System - Full conversation tracking with analysis tools
• Automatic Parallelization - Smart detection and execution of parallel tasks

Community Fork Notice

This is a community-maintained fork of Google's Gemini CLI. It is not affiliated with, endorsed by, or supported by Google or Alibaba Cloud.

• Original Project: google-gemini/gemini-cli by Google
• This Fork: Community project for Qwen model integration
• Support: Community-based support through GitHub issues
• License: Apache License 2.0 (same as original)

For the original project's terms of service and privacy notice, see the original repository.

Contributing to This Fork

We welcome contributions to improve Qwen model integration and CLI functionality. Please see CONTRIBUTING.md for guidelines specific to this community fork.