xerrors/Yuxi

Yuxi is a multi-tenant agent harness platform that integrates knowledge base management with knowledge graph capabilities, built on LangChain, Vue, and FastAPI. The system enables enterprises to make their knowledge retrievable, reasoned over, and deliverable by intelligent agents. It combines RAG retrieval, knowledge graphs stored in Milvus, and multi-agent orchestration through LangGraph into a unified multi-tenant workspace where administrators configure knowledge bases, models, and permissions while users interact with agents through a ChatGPT-like interface.

The platform's core architecture supports agents that can mount Skills, MCP tools, sub-agents, and sandbox utilities. Users receive answers with cited sources, knowledge graph reasoning, and deliverable outputs. The technical stack spans Vue 3 with Vite and Pinia on the frontend, FastAPI with LangGraph and ARQ async workers on the backend, and PostgreSQL, Redis, MinIO, Milvus, and Neo4j for storage. Document parsing is handled through MinerU, PaddleX, and RapidOCR, with Docker Compose enabling deployment.

The repository is classified across multiple domains including Knowledge Graph, Information Extraction, Natural Language Processing, Data Mining, Machine Learning, Data Visualization, Graph Database, Smart City, Semantic Web, and AI Applications. This broad classification reflects the platform's comprehensive approach to enterprise knowledge management and intelligent agent development.

Yuxi supports multiple advanced features including DeepAgents as a deep agent framework, MinerU for PDF processing, Neo4j for graph database operations, and MCP for tool integration. The platform offers both standard and lightweight deployment modes, with the LITE mode available for scenarios where heavy dependencies like knowledge bases and knowledge graphs are not required. The project provides comprehensive documentation, version-specific features, and a development roadmap, with the latest updates tracked in a detailed changelog. The platform is licensed under MIT, making it freely available for both commercial and personal use.