agentscope: Building Distributed LLM Agent Systems

agentscope is a Python-based MCP Client designed to streamline the creation of multi-agent applications powered by large language models (LLMs). It provides a framework that emphasizes ease of use, robustness, and distributed execution, making it suitable for developers looking to build and deploy complex agent-based systems.

Simplified Development with Workstation and Copilot

One of agentscope's standout features is its AgentScope Workstation. This component offers a drag-and-drop programming platform, significantly lowering the barrier to entry for building multi-agent applications. For new users, a built-in copilot further assists in the development process, guiding them through the framework's capabilities. This visual approach complements its comprehensive documentation and fruitful components, aiming to make agent system development more accessible.

Robustness and Distributed Execution

agentscope is built with high robustness in mind, incorporating customized fault-tolerance controls and retry mechanisms. These features are crucial for enhancing the stability of LLM-powered applications, which can sometimes be prone to transient errors.

For scaling and performance, agentscope leverages an actor-based distribution model. This allows developers to construct distributed multi-agent applications while maintaining a centralized programming paradigm. This approach simplifies the management of complex, concurrently running agent systems, abstracting away much of the underlying distribution complexity.

Broad LLM Compatibility

The framework offers extensive compatibility with various LLM services and local deployment options through its ModelWrapper. This includes support for popular third-party APIs such as OpenAI, DashScope, Gemini, ZhipuAI, ollama, LiteLLM, Yi API, and Anthropic API.

For developers preferring local LLM deployment, agentscope integrates with libraries like ollama, Flask, Transformers, ModelScope, FastChat, and vllm. This flexibility ensures that agentscope can be integrated into diverse development environments and leverage a wide array of LLM resources.

Practical Applications

agentscope supports a range of example applications, demonstrating its versatility. These include basic and self-organizing conversations, integration with LangChain for extended capabilities, and implementations of ReAct agents. More complex scenarios like natural language to SQL queries, RAG (Retrieval Augmented Generation) agents, and even games like Gomoku and Werewolf showcase the framework's ability to handle diverse agent interactions and problem-solving tasks. The AgentScope Studio further provides a runtime user interface for displaying multimodal output, including text, images, audio, and video.

Setup and Installation

agentscope requires Python 3.9 or higher. For active development, installing from source is recommended:

# Install the package in editable mode
cd agentscope
pip install -e .

Alternatively, it can be installed via pip:

pip install agentscope

Optional dependencies for specific deployment scenarios, such as distribution mode, can be installed as needed.

agentscope: LLM Multi-Agent Application Framework MCP Client