Sage is a production-ready, modular multi-agent orchestration framework designed to simplify complex problem-solving. It targets developers and researchers needing to break down intricate tasks into manageable subtasks through intelligent agent collaboration, offering both deep analysis and rapid execution modes.

How It Works

Sage employs an agent-based architecture where a central controller orchestrates specialized agents (Analysis, Planning, Execution, Observation, Summary) to tackle complex tasks. It features a robust tool system with MCP server support and auto-discovery, enabling agents to leverage external functionalities. The framework prioritizes extensibility, error handling, and efficient token management through its Message Manager.

Quick Start & Requirements

• Install: git clone https://github.com/ZHangZHengEric/Sage.git, cd Sage, pip install -r requirements.txt, pip install -r examples/fastapi_react_demo/requirements.txt
• Prerequisites: Python 3.10+, OpenAI API key (or other supported providers).
• Demo: Live demo available at http://36.133.44.114:20040/.
• Docs: Quick Start Guide, Architecture Overview.

Highlighted Details

• Supports a wide range of LLMs including GPT-4, Claude 3.5, Qwen, and DeepSeek-V3 via direct API or OpenRouter.
• Features an interactive web interface (React + FastAPI) with real-time streaming visualization, workflow management, and response interruption.
• Includes advanced token tracking and cost monitoring, with a Message Manager designed for smart filtering and compression.
• Offers a custom workflow engine with a visual mind-map editor and a rule preferences system for personalized AI behavior.

Maintenance & Community

The project is actively maintained, with recent updates (v0.9.2) including UI enhancements and bug fixes. Links to documentation and examples are provided within the README.

Licensing & Compatibility

Licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The live demo URL may not always be stable. While extensive, the list of officially tested models might not cover all potential integrations.