Alibaba's AI lab Qwen recently introduced a new framework designed to help developers create what they call AI agents.
Qwen-Agent builds on Qwen's existing language models, giving these agents essential capabilities like following detailed instructions, using tools, planning tasks, and maintaining conversation context. The framework also comes with advanced features including RAG (retrieval-augmented generation), a code interpreter, and specialized mathematical reasoning powered by Qwen2.5-Math.
The framework takes a two-tiered approach to agent development. The base layer provides language models and basic tools, while the top layer provides ready-to-use agent components. Developers can combine these pieces to build agents that can perform complex tasks - from reading PDFs to working with existing tools to performing custom functions. A real-world example is BrowserQwen, a web browsing agent/assistant that shows what the framework can do.
For implementation, developers can either use Alibaba's DashScope cloud service or run Qwen models on their own hardware. Alibaba recently slashed prices for its API AI services. The framework also includes a GUI that makes it easy to create interactive web demos using the Gradio framework.
Alibaba keeps expanding the technology behind these agents. They've just released QVQ-72B-Preview for visual tasks, alongside specialized Qwen2.5 models for programming and mathematics.
However, developers should keep two things in mind: like other Chinese LLMs, these agents may have restrictions around political content, and it's worth considering simpler solutions before jumping into agent-based approaches.
