Yuxi

Yuxi is a self-hosted, multi-tenant knowledge agent platform that unifies retrieval-augmented generation, knowledge graphs and multi-agent workflows in one deployable system. Teams can feed it their documents and then question, explore and build agents over that knowledge from a web interface. The project has about 7,400 stars on GitHub, is released under the MIT license, and was actively developed in 2026.

Overview

Yuxi is built with FastAPI and Vue, and combines a Milvus-powered vector store for RAG with a Neo4j knowledge graph, so answers can draw on both semantic search and explicit relationships between entities. On top of that knowledge layer sit agents: LangGraph- and DeepAgents-based workflows that can use MCP tools and skills, run inside a sandbox, and operate under a permission system designed for multi-tenant teams. The result is something between a private ChatGPT-over-your-documents and an internal agent platform — an organization deploys it once, connects its knowledge bases, and gives staff a governed way to chat with documents, run knowledge-base QA and automate research tasks without sending data to external SaaS tools.

Key Features

• Unified RAG over documents with Milvus vector search
• Knowledge graph layer (Neo4j) for entity and relationship-aware answers
• Multi-agent workflows built on LangGraph and DeepAgents
• MCP and skills support for connecting external tools
• Multi-tenant design with sandboxing and permission management
• Self-hosted with Docker, using a FastAPI backend and Vue frontend
• Document ingestion pipeline that chunks, embeds and indexes files automatically
• Web admin console for managing knowledge bases, agents and users

💳 Pricing

Free. Yuxi is open source under the MIT license for self-hosting; you cover your own infrastructure and model API costs.

Who Is It For

Enterprises, research groups and internal platform teams that want a private, knowledge-grounded AI assistant — document QA, knowledge exploration and agent workflows — under their own control.