Yuxi

This entry covers Yuxi, the project published by the xerrors account. Yuxi is a multi-tenant knowledge agent platform that organizations can deploy privately, for teams that want to control their own data, models and permissions. It brings knowledge base retrieval, knowledge graphs, LangGraph multi-agent orchestration, MCP and Skills, sandboxed tools and permission management into one workspace, with the tagline of giving every user a cloud agent.

Capabilities include customizable agents, an agentic RAG toolkit for uploading, parsing, chunking and indexing documents, multi-step task execution that combines tools, MCP, Skills, sub-agents and sandboxes to produce downloadable files, and knowledge graph building that extracts entities and relationships from Milvus document chunks into Neo4j. Admins can manage knowledge bases, agents, Skills and models by user, department and sharing scope, and quality can be compared through retrieval evaluation and Langfuse datasets.

Yuxi runs entirely on Docker, with services including LangGraph, Vue, FastAPI, Milvus, Neo4j, PostgreSQL, MinerU and PaddleOCR, and needs a model API. It is MIT licensed and written in Python, and the documentation, mostly in Chinese, covers quick start, model setup, knowledge bases and production deployment.