Qwen-Agent

Stale
GitHub Python Apache-2.0

Description

Agent framework built on Qwen LLM, featuring function calling, MCP tool integration, code interpreter, RAG, and browser extension support.

Key Features

  • Built on the Qwen family of LLMs, supporting Qwen2.5, QwQ-32B, Qwen3, and the latest Qwen3.5 models
  • Built-in function calling with parallel, multi-step, and multi-turn tool calls, compatible with vLLM and DashScope API
  • Supports MCP (Model Context Protocol) for seamless integration with external tools and data sources
  • Ships with ready-to-use example applications including Code Interpreter, RAG, and Browser Assistant
  • Offers atomic components (LLM, Tool) and high-level Agent abstractions for rapid custom agent development
  • Integrates Gradio 5 GUI for visual debugging and interactive development

Use Cases

💡 Building intelligent assistants with tool-calling capabilities for complex multi-step task automation
💡 Developing private agent applications based on Qwen models with local deployment support
💡 Building enterprise-grade RAG systems for document retrieval and intelligent Q&A
💡 Leveraging the code interpreter for data analysis and mathematical reasoning tasks
💡 Integrating third-party tools via MCP protocol to extend agent's external interaction capabilities

Strengths & Limitations

Strengths

  • High community interest (17.1k stars)
  • Permissive open-source license (Apache-2.0)
  • Established track record (2 years in production)

⚠️ Limitations

  • No updates in over 6 months

Quick Start

Install: pip install -U 'qwen-agent[gui,rag,code_interpreter,mcp]'; set DASHSCOPE_API_KEY env var or deploy locally with vLLM/Ollama; create LLM config and register custom tools; instantiate Assistant Agent and call agent.run() to execute tasks.

Related Projects

fast-agent

3.9k · Python
Active A+

Code, Build and Evaluate agents - excellent Model and Skills/MCP/ACP Support.

agent-frameworkmcpskills +1
  • · CLI-first interactive agent framework with prompt_toolkit-powered TUI, shell completions, and streaming responses via Rich
  • · Complete MCP feature support including Sampling and Elicitations — the first framework with end-to-end tested MCP integration
  • · Declarative agent definition with minimal boilerplate — define agents via @fast.agent decorator and compose prompts with MCP servers

FIM One

1.5k · Python
Active A

Open-source agent platform for Global × China enterprises — wire every system through one agent core. Self-hosted, any LLM.

enterprisemcprag +2
  • · All-in-one agent platform — Standalone assistant, embedded Copilot, or central Hub in one agent core
  • · Cross-border system connectivity — connects global SaaS and China stack (Feishu, WeCom, DingTalk, DM, Kingbase) via one pattern
  • · Dynamic DAG planning — LLM decomposes goals into dependency graphs at runtime with parallel execution

LangChain

145.5k · Python
Active A+

LangChain is the open-source agent engineering platform that unifies model IO, tool calling, RAG, memory and observability under one composable framework.

agent-frameworkragorchestration +2
  • · Unified model abstraction — ChatModel interface covers OpenAI, Anthropic, Bedrock, Vertex and 100+ providers
  • · Tool calling & agents — tool calling, ReAct, OpenAI Tools, Plan-and-Execute agent paradigms
  • · RAG as a first-class citizen — Document Loader, Text Splitter, Retriever and Index abstractions

Related Articles