1. Project Overview

AgentScope: Agent-Oriented Programming for Building LLM Applications

Start building LLM-empowered multi-agent applications in an easier way.

Open-source multi-agent development platform, helping developers easily build and deploy multi-agent applications.

AgentScope provides high usability, high robustness, and distributed support, with multiple model APIs and local model deployment options, covering chat, image synthesis, text embedding, etc. multiple tasks.

AgentScope includes drag-and-drop programming interface, interactive programming assistant, real-time monitoring function, and rich development resources, supporting rapid secondary development.

AgentScope has fault-tolerant mechanisms, prompt optimization, and distributed parallel processing capabilities, improving application performance and development efficiency.

2. Open Source License

Licensed under Apache-2.0

3. Screenshots

4. Feature Overview

✨ Why AgentScope？

Easy to get started, profound in use.

• Transparent to developers
  Transparency is the primary principle of AgentScope. Regardless of prompt engineering, API calls, intelligent entity construction, or workflow orchestration, we adhere to being transparent and controllable to developers. We reject deep encapsulation or implicit magic.
• Supports model-agnostic development
  Write once, adapt to all models. Supports more than 17+ 
• different LLM APIs.
  ‘Lego-style’ intelligent entity construction All components remainmodular and mutually independent, with the developer having the final say on whether to use them or not.
• Multi-Agent
  Designed for Multi-AgentDesign, Explicitmessage passing and workflow orchestration, reject deep encapsulation.
• native distributed/parallelization
  native support for distributed intelligent agent applications, support Auto-parallelizationand Centralized Orchestration。
• Highly Customizable
  Tools, Tutorials, Agents, Workflow Orchestration, Visualization, AgentScope supports & encourages developers to customize.
• Developer-friendly
  Low-code development, Visualized tracking and monitoring. From development to deployment, one-stop solution.

5. Tech Stack

Quick Start

AgentScope requires Python 3.9 or higher. You can install from source or pypi.

Install from PyPI

pip install agentscope

Install from Source

To install AgentScope from source, you need to clone the repository from GitHub and then install using the following command:

git clone agentscope
cd agentscope
pip install -e .

To ensure AgentScope is installed correctly, you can run the following code:

import agentscope

print(agentscope.__version__)

0.1.6

Extra Dependencies

AgentScope provides extra dependencies for different needs. You can install them as needed.

• ollama: Ollama API
• litellm: Litellm API
• zhipuai: Zhipuai API
• gemini: Gemini API
• anthropic: Anthropic API
• service: dependency for tool functions
• distribute: dependency for distributed mode
• full: install all dependencies at once

For example, in distributed mode, the installation command varies by operating system.

For Windows users:

pip install agentscope[gemini]\n# or\npip install agentscope[ollama,distribute]

For Mac and Linux users:

pip install agentscope[gemini]
# or
pip install agentscope[ollama,distribute]

Total running time of the script: (0 minutes 2.370 seconds)