muAgent

One-Minute Overview#

muAgent is a novel agent framework driven by both Large Language Models and EKG (Eventic Knowledge Graph, Industry Knowledge Carrier), enabling complex SOP execution through human-machine collaboration via canvas-based drag-and-drop and simple text writing. It's compatible with existing market frameworks and offers four core differentiated technical functions: Complex Reasoning, Online Collaboration, Human Interaction, and Knowledge On-demand. Ideal for enterprise development teams requiring complex workflow automation, team collaboration, and knowledge management.

Core Value: Significantly improves efficiency and accuracy in complex process execution through knowledge graph-enhanced agent collaboration.

Quick Start#

Installation Difficulty: Medium - Requires Docker environment and basic AI model configuration

# Using EKG services in three steps
# Step 1: Clone the repository
git clone https://github.com/codefuse-ai/CodeFuse-muAgent.git

# Step 2: Navigate to the directory
cd CodeFuse-muAgent

# Step 3: Start all container services, this might take some time
docker-compose up -d

Is this suitable for me?

• ✅ Enterprise DevOps Scenarios: Validated in multiple complex DevOps scenarios at Ant Group
• ✅ Team Collaboration: Supports multiplayer text-based games and collaborative workflows
• ✅ Knowledge Management: Supports intelligent parsing and one-click import of extensive documentation
• ❌ Simple Task Automation: May be overly complex for straightforward processes
• ❌ Resource-Constrained Environments: Requires Docker setup and LLM services

Core Capabilities#

1. EKG Builder - Solving Knowledge System Construction Challenges#

Through virtual team design, scenario intentions, and semantic nodes, you can experience differences between online/local documentation and annotated/unannotated code handovers. For large amounts of existing documents (text, diagrams, etc.), we support intelligent parsing with one-click import capability. Actual Value: Significantly reduces knowledge system construction costs and improves team handover efficiency

2. EKG Assets - Solving Complex SOP Automation Requirements#

Through comprehensive KG Schema design—including Intention Nodes, Workflow Nodes, Tool Nodes, and Character Nodes—we can meet various SOP automation requirements. Tool Nodes enhance accuracy in tool selection and parameter filling, while Character Nodes enable human-involved process advancement. Actual Value: Flexibly adapts to various automation scenarios, supporting everything from fully automated to human-machine collaborative workflows

3. EKG Reasoning - Solving Unknown Scenario Exploration#

Unlike purely model-based or entirely fixed-flow reasoning, our framework allows LLMs to operate under human guidance, providing flexibility, control, and enabling exploration in unknown scenarios. Successful exploration experiences can be summarized and documented into KG, minimizing detours for similar issues. Actual Value: Improves adaptability and problem-solving efficiency in unknown scenarios while maintaining control

4. Diagnosis - Solving Process Debugging Challenges#

After KG editing, the visual interface enables quick debugging, and successful execution path configurations are automatically documented, reducing model interactions, accelerating inference, and minimizing LLM token costs. We also provide comprehensive end-to-end visual monitoring during online execution. Actual Value: Dramatically reduces debugging time and improves execution efficiency and transparency

5. Memory - Solving Long Context Processing Issues#

The unified message pooling design supports categorized message delivery and subscription based on different scenario needs, such as multi-agent environments. Through message retrieval, reranking, and distillation, it facilitates long-context processing and improves overall Q&A quality. Actual Value: Effectively handles long conversation scenarios, improving response accuracy and consistency

6. Action Space - Solving Tool Calling and Code Execution Challenges#

Following the Swagger protocol, we provide tool registration, categorization, and permission management to facilitate LLM function calling. We offer a secure and trustworthy code execution environment, ensuring precise code generation for various scenarios including visual plotting, numerical calculations, and table editing. Actual Value: Ensures accuracy in tool calling and safety in code execution, expanding the system's application scope

Technology Stack & Integration#

Development Languages: Python, Shell Key Dependencies: Docker, Docker Compose, LLM services Integration Method: SDK / Containerized services

Maintenance Status#

• Development Activity: Actively developed - Recently released v2.0 with ongoing updates
• Recent Updates: Recent - Released v2.0 on September 5, 2024
• Community Response: Responsive - Actively engages with user feedback through GitHub Issues

Documentation & Learning Resources#

• Documentation Quality: Basic - Has documentation and examples, but limited API documentation
• Official Documentation: https://github.com/codefuse-ai/CodeFuse-muAgent
• Example Code: Available - Documentation mentions ~/examples directory contains examples