agentscope: LLM Multi-Agent Application Framework and MCP Client
agentscope: Building Distributed LLM Multi-Agent Systems
agentscope is a Python-based MCP Client designed to streamline the creation of LLM-powered multi-agent applications. It provides a framework that simplifies the complexities of distributed agent systems, offering tools like a drag-and-drop workstation and robust fault-tolerance mechanisms.
Simplifying Multi-Agent Development
The core of agentscope lies in its approach to multi-agent application development. It uses an actor-based distribution model, allowing developers to build distributed systems with a centralized programming paradigm. This abstraction aims to reduce the overhead typically associated with managing distributed components.
Key features supporting this include:
- Fruitful Components: A collection of pre-built elements to accelerate development.
- High Robustness: Custom fault-tolerance controls and retry mechanisms are built-in to enhance application stability.
- Comprehensive Documentation: Resources to guide developers through the framework.
- Broad Compatibility: Support for various model libraries and local deployment options.
Bridging to Diverse LLMs
agentscope offers extensive compatibility with a range of LLM providers and local deployment methods through its ModelWrapper. This allows developers to integrate preferred models without significant refactoring.
Supported third-party APIs include:
- OpenAI
- DashScope
- Gemini
- ZhipuAI
- ollama
- LiteLLM
- Yi API
- Anthropic API
For local model deployment, agentscope integrates with libraries such as:
- ollama
- Flask
- Transformers
- ModelScope
- FastChat
- vllm
This flexibility ensures that applications built with agentscope can leverage both cloud-based and self-hosted LLM solutions.
The AgentScope Workstation and Studio
For developers new to multi-agent systems or looking for a visual approach, agentscope provides two key interfaces:
- AgentScope Workstation: A drag-and-drop programming platform that includes a copilot feature, assisting beginners in constructing agent-based systems.
- AgentScope Studio: An easy-to-use runtime user interface capable of displaying multimodal output, including text, images, audio, and video, on the front end. This provides a visual way to monitor and interact with running agent applications.
Additionally, agentscope includes a Web Search tool, allowing agents within the system to retrieve information from the internet.
Getting Started
agentscope requires Python 3.9 or higher. While a pip install agentscope command is available, the project is under active development, and installing from source in editable mode is recommended:
cd agentscope
pip install -e .Optional dependencies for specific deployment scenarios, such as distribution mode, can be installed as needed.