AgentUniverse: A Multi-Agent Collaboration Framework for Complex Business Scenarios 126

What is agentUniverse?

agentUniverse is a multi-agent collaboration framework designed for complex business scenarios, originating from Ant Group's real-world financial business practices.

It provides quick and easy-to-use capabilities for building large model agent applications, as well as a rich set of multi-agent collaboration mode components (such as PEER and DOE modes), helping developers and enterprises build domain-expert-level agents that work collaboratively to complete tasks.

agentUniverse Features

Flexible and Extensible Agent Building

• Provides key components required for agent building, supporting customized development based on user needs.

• Supports enhancing agent capabilities through tools, knowledge bases, RAG technology, etc.

Multi-Agent Collaboration Modes

PEER Mode:

• Through four agents with different responsibilities—Plan, Execute, Express, and Review—multi-step decomposition and step-by-step execution of complex problems are achieved, with autonomous iteration based on evaluation feedback.

• Suitable for scenarios requiring deep analysis, such as event interpretation and industry analysis.

DOE Mode:

• Through three agents—Data-fining, Opinion-inject, and Express—the effectiveness of data-intensive, high-computational-precision generation tasks that integrate expert opinions is enhanced.

• Suitable for scenarios such as financial report generation.

• Supports users to define and orchestrate new collaboration modes, enabling multiple agents to work together organically.

Domain Experience Integration

• Provides capabilities for domain prompts and knowledge construction and management.

• Supports domain-level SOP orchestration and injection, helping agents reach a domain-expert level.

Development Support

• Standard Project Scaffold: Helps developers quickly build single-agent or multi-agent applications.

• Canvas-based R&D Platform: Provides a local canvas-based development platform that supports one-click startup.

Application Practice

Investment Research Assistant: This is an investment research assistant built on agentUniverse, designed to improve the efficiency of investment research experts. It has been validated across multiple large-scale scenarios at Ant Group.

Agent Applications: agentUniverse can be used to build various agent applications, covering fields such as investment research, ESG, finance, and financial reporting.

How to Use

1. Quick Installation

Install via the pip command:

pip install agentUniverse

2. Building an Agent Application

Standard Project Scaffold:

• Quickly Build a Single Agent: Read the "Quickly Build a Single Agent" section to learn how to build a single agent and master methods for enhancing agent capabilities.

• Build a Multi-Agent Application: Read the "Build a Typical Multi-Agent Application" section to learn how to split tasks into multiple agents working collaboratively.

Canvas-based R&D Platform:

• Install dependencies:

pip install magent-ui ruamel.yaml

• One-click run:

Run the product_application.py file under sample_apps/workflow_agent_app/bootstrap/platform to launch the local canvas-based R&D platform.

GitHub Repository: https://github.com/agentuniverse-ai/agentUniverse