Yuxi
可私有部署的多租户知识智能体平台:统一 RAG、知识图谱、多智能体、MCP/Skills、沙盒与权限管理。Yuxi = Cloud Agents + Knowledge RAG, Self-hosted knowledge agent pla
What is Yuxi?
可私有部署的多租户知识智能体平台:统一 RAG、知识图谱、多智能体、MCP/Skills、沙盒与权限管理。Yuxi = Cloud Agents + Knowledge RAG, Self-hosted knowledge agent platform for RAG, knowledge graphs and multi-agent workflows.
Overview of Yuxi
What it is
Yuxi is an open-source, self-hosted multi-tenant knowledge agent platform that unifies RAG, knowledge graphs, multi-agent workflows, MCP/Skills, sandbox tools, and permission management in a single workspace. It is designed for teams that need to deploy their own multi-user agent product while retaining full control over data, models, and permissions.
Key features
- Unified agent workspace with multi-turn dialogue, knowledge base retrieval, task status tracking, and human approval.
- Knowledge base and RAG suite supporting document upload, parsing, chunking, vector indexing, and retrieval testing with rerank.
- Knowledge graph extraction from document chunks stored in Milvus, with entity and relation storage in Neo4j for graph-based retrieval.
- Multi-agent orchestration with LangGraph, including sub-agents for parallel task execution, Skills plugin system, and MCP (Model Context Protocol) standard protocol support.
- Sandbox workspace with isolated file system, file generation, online preview, and download capabilities.
- Team governance with multi-tenant management, user/department permissions, model configuration, API key management, and usage dashboards.
Use cases
Yuxi is suitable for teams that need to build custom AI agents with internal knowledge bases, execute multi-step tasks combining tools and external services, manage team access to knowledge, agents, and models, and evaluate retrieval and agent performance using built-in assessment tools.
Getting started
bashgit clone --branch v0.7.3 --depth 1 https://github.com/xerrors/Yuxi.gitcd Yuxi./scripts/init.shdocker compose up --build -dAfter startup, check readiness with curl --fail http://localhost:5050/api/system/ready. Once the status is ready, open http://localhost:5173 and follow the prompts to initialize the super admin and log in.
License and pricing
MIT license. Open source.
Based on the project's public README and repository metadata.