Modelscope-Agent

It is an agent framework called ModelScope-Agent that connects models within the ModelScope platform to external environments, enabling customizable and scalable agent systems. The framework allows a single agent to perform tasks such as role-playing, calling large language models (LLMs), using tools, planning, and memory management. It is designed to integrate LLMs with task-specific tools, where the LLM determines which tools to invoke to complete user tasks.

To use the framework, users initialize a RolePlay object with specific tasks. The Agent object consists of components like message/prompt generation, LLM calls, and tool invocation based on LLM results. The LLM module is central to the agent, ensuring high-quality interaction outcomes. The framework supports multiple model service providers, including Dashscope (for Qwen series models), Zhipu (for GLM series models), and OpenAI (for OpenAI API-compatible models). Users can also customize their LLMs by implementing specific functions like `_chat_stream` and `_chat_no_stream`.

The framework provides built-in multi-domain tools that can be configured for use within the agent. Users can also create custom tools by defining the tool’s name, description, and parameters, and implementing the `call()` function. An example of custom tool creation is provided in the `demo_register_new_tool` section.

The ModelScope-Agent framework is open-source and licensed under the Apache License (Version 2.0). It integrates with projects like FaceChain, an open-source tool for generating personalized portraits, simplifying the process through dialogue with the FaceChain Agent. Users can run the framework in environments like ModelScope Notebook, Colab, or local setups. For further assistance, users can contact the team via email or participate in the open-source community by reporting issues or sharing their work through pull requests.