Wenjie ZhangYuxi
A self-hosted, multi-tenant knowledge agent platform
Bring RAG, knowledge graphs, and multi-agent execution into one workspace
[Project home] · [Quick start] · [Demo] · [Releases] · [中文]
Introduction
Yuxi is a self-hosted, multi-tenant knowledge agent platform. Rather than stopping at a chat interface, it brings RAG retrieval, Milvus-backed knowledge graphs, LangGraph multi-agent orchestration, MCP/Skills, sandbox tools, and access control into one workspace.
Administrators connect model providers, build knowledge bases, and manage user or department permissions. Users work with agents that can retrieve cited sources, reason over graph context, call tools and sub-agents, and deliver previewable, downloadable artifacts.
Navigation: Introduction ｜ Quick Start ｜ Roadmap; for the latest updates, see the changelog.
Core Features
- 🤖 Agent development — Built on LangGraph, with sub-agents (SubAgents), Skills, MCPs, Tools, and middleware; long-running tasks run asynchronously on a background worker, backed by a sandbox file system for persisting, previewing, and downloading tool artifacts.
- 📚 Knowledge base (RAG) — Multi-format document parsing (MinerU / PaddleX / OCR), configurable Embedding and Rerank models, knowledge base evaluation, in-app PDF / image preview, and retrieval sources backfilled as chat citations.
- 🕸️ Knowledge graph — Build, visualize, and retrieve entity-relation graphs inside Milvus knowledge bases, then fuse graph hits with chunk retrieval for agent reasoning.
- 🏢 Multi-tenancy & permissions — User / department-level access control, unified model provider configuration, and API Key authentication for external system integration.
- ⚙️ Platform & engineering — Vue + FastAPI architecture, ready-to-run Docker Compose deployment, dark mode, a lightweight LITE startup mode, and production-grade orchestration.
When Yuxi Fits
Yuxi is a strong fit for teams that need private deployment, organizational access control, multiple knowledge sources, and extensible agents that can execute work. If you only need a minimal single-document chat UI or a fully managed SaaS with no infrastructure to operate, Yuxi may be more platform than you need.
Tech Stack
| Layer | Technologies |
|---|---|
| Frontend | Vue 3 · Vite · Pinia |
| Backend | FastAPI · LangGraph · ARQ (async worker) |
| Storage | PostgreSQL · Redis · MinIO · Milvus · Neo4j |
| Doc parsing | MinerU · PaddleX · RapidOCR |
| Deployment | Docker Compose |
Quick Start
Prerequisites: Docker and Docker Compose installed, plus at least one OpenAI-compatible LLM API.
1. Clone and initialize
git clone --branch v0.7.1 --depth 1 https://github.com/xerrors/Yuxi.git
cd Yuxi
# Linux/macOS
./scripts/init.sh
# Windows PowerShell
.\scripts\init.ps1
2. Start with Docker
docker compose up --build
Do not run up directly when upgrading an existing installation to the v0.7.2
storage layout. The single owning procedure is the
production deployment guide.
3. Open the platform
Once the services are ready, open http://localhost:5173 in your browser and sign in with the admin account generated during initialization.
💡 If you don't need heavy dependencies like knowledge bases / graphs, run
make up-litefor a lightweight LITE mode with faster cold starts. See the docs for more deployment details.
Examples and Demo
Home |
Dashboard Statistics |
Agent Configuration |
Knowledge Base Invocation |
|
Create Knowledge Base |
Knowledge Base Management |
|
Knowledge Graph Visualization |
Project Documentation |
Extension Management (Skills) |
Extension Management (MCPs) |
|
User / Department Permission Management |
Model Provider Configuration |
Acknowledgements
Yuxi references and builds on the following excellent open-source projects:
- LightRAG - Inspired parts of the early graph construction and retrieval design. Yuxi now uses its own Milvus-backed knowledge-base and graph pipeline.
- DeepAgents - Used as the deep agent framework.
- DeerFlow - Referenced for Sandbox agent architecture ideas.
- RAGflow - Referenced for document text chunking strategies.
- LangGraph - Multi-agent orchestration framework and the core architectural foundation of this project.
- QwenPaw - Referenced for model configuration and personal file area design.
Contributing
Thanks to all contributors for supporting this project!
Star History
📄 License
This project is licensed under the MIT License. See LICENSE for details.
Third-party components brought in by Docker Compose (Neo4j Community GPL-3.0, MinIO AGPL-3.0, etc.) retain their original licenses; see the deployment guide for deployment and redistribution boundaries.
If this project helps you, please give us a ⭐️.