Awesome AI Agents › Knowledge Graph Orchestration

codefuse-ai/CodeFuse-muAgent

⭐ 773 Python repository created 2024-04-21

muAgent is an innovative agent framework driven by a Knowledge Graph (KG) engine, designed to enhance multi-agent orchestration and collaboration. It integrates large language models (LLM) with an Eventic Knowledge Graph (EKG), which serves as an industry knowledge carrier, enabling complex reasoning, online collaboration, human interaction, and knowledge on-demand. The framework supports multi-agent systems, function calling, and code interpretation, allowing users to execute complex standard operating procedures (SOPs) through a user-friendly canvas-based drag-and-drop interface or simple text commands under human guidance. muAgent is compatible with existing frameworks and has been validated in complex DevOps scenarios, notably within Ant Group. The framework offers several key features: EKG Builder allows the design of virtual teams, scene intentions, and semantic nodes, supporting intelligent parsing and one-click import of large document sets. EKG Assets include comprehensive KG schema design with intention nodes, workflow nodes, tool nodes, and character nodes, facilitating SOP automation and human-involved process advancement. EKG Reasoning combines model-based and fixed-flow reasoning, enabling LLMs to operate flexibly under human guidance and document successful explorations to minimize detours. The Diagnose feature provides a visual interface for debugging and execution path documentation, reducing model interactions and inference time while offering end-to-end visual monitoring during execution. Memory management supports categorized message delivery and long-context handling to improve question-answer quality. ActionSpace adheres to the Swagger protocol for tool registration, categorization, permission management, and secure code execution, supporting various scenarios like visual plotting, numerical calculations, and table editing. muAgent is open source and provides an SDK for easy integration. It supports one-click deployment with Docker containers and offers extensive documentation and examples. The project encourages community contributions and provides contact options for support and collaboration. Overall, muAgent is a powerful framework for building intelligent, collaborative multi-agent systems driven by knowledge graphs and large language models, suitable for complex automation and interactive applications.

https://github.com/codefuse-ai/CodeFuse-muAgent

agentagent-frameworkcode-interpretercomplex-reasoningdevopsdockerdrag-and-drop-interfaceekgeventic-knowledge-graphfunction-callinghuman-interactionkgkg-schemaknowledge-graphknowledge-on-demandlarge-language-modelllmmemory-managementmulti-agentmulti-agent-orchestrationonline-collaborationopen-sourceragsdksecure-code-executionsemantic-nodessop-automationtool-learningtool-registrationvirtual-teams