AgentScope is a multi-agent platform designed to simplify the development of LLM-powered applications. It targets developers seeking to build complex agentic systems, offering features for easy integration, robust fault tolerance, and streamlined distributed development.

How It Works

AgentScope employs an actor-based distribution model, enabling centralized programming for distributed multi-agent systems. It decouples model invocation from deployment using a ModelWrapper abstraction, supporting a wide array of LLM providers (OpenAI, DashScope, Gemini, Ollama, etc.) and local model deployments (Flask, FastChat, vLLM). This approach allows for flexible model selection and configuration, enhancing application robustness with customizable fault-tolerance and retry mechanisms.

Quick Start & Requirements

• Install from source: git clone https://github.com/modelscope/agentscope.git && cd agentscope && pip install -e .
• Install via pip: pip install agentscope
• Requires Python 3.9+. Optional dependencies for specific features (e.g., [distribute]) are available.
• Official documentation and tutorials are available.

Highlighted Details

• Supports a broad range of LLM APIs and local model services.
• Features an "AgentScope Workstation" for drag-and-drop application building.
• Achieved a 63.4% resolve rate on SWE-Bench(Verified).
• Integrates tools, web browser control, and multimodal capabilities.

Maintenance & Community

• Active development with recent updates including hook functions, tools API, and MCP Server support.
• Community channels available via Discord and DingTalk.
• Roadmap is publicly available.

Licensing & Compatibility

• Released under Apache License 2.0.
• Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

The project is in active development, with version 0.1.0 released in late 2024. While robust features are highlighted, users should anticipate potential changes and evolving API stability.