Verified project record

xerrors/Yuxi

This project provides a unified, multi-tenant workbench for building and querying enterprise knowledge bases using AI agents. It integrates RAG retrieval, Milvus knowledge graphs, and LangGraph multi-agent orchestration, accessible via a conversational web interface.

Project overview

Project type: AI Agent · RAG · Workflow
Use cases: Knowledge Q&A
Deployment: Refer to project documentation
License: MIT

Best for

Enterprise teams needing a self-hosted, multi-tenant platform to deploy configurable AI agents with integrated RAG and knowledge graph reasoning.
Developers looking to orchestrate multi-agent workflows using LangGraph and interact with enterprise knowledge bases via a conversational interface.

Key capabilities

Provides a unified platform where administrators can configure knowledge bases, models, and permissions for different tenants.
Uses Retrieval-Augmented Generation to allow users to search and retrieve information from enterprise knowledge bases.
Uses LangGraph to coordinate multiple intelligent agents for complex task execution and reasoning.
Users interact with agents via a conversational interface, receiving answers with cited sources and knowledge graph reasoning.
Supports document parsing using MinerU, PaddleX, and RapidOCR for ingesting knowledge base content.
Agents can be configured with attachable Skills, MCP (Model Context Protocol), sub-agents, and sandbox tools.

Limitations and risks

The platform requires a paid external service in the form of an OpenAI-compatible LLM API to function, meaning it cannot operate fully offline without external dependencies.
GPU requirements, minimum hardware specifications, and telemetry practices are not documented.

Getting started

Ensure Docker and Docker Compose are installed. Clone the repository, run the initialization script, execute 'docker compose up --build', and open http://localhost:5173 in a browser.
Prepare at least one OpenAI-compatible LLM API to enable language model capabilities for the agents.

Evidence and sources

GitHub project description: 结合知识库、知识图谱管理的多租户 Agent Harness 平台。 An agent harness that integrates a LightRAG knowledge base and knowledge graphs. Build with LangChain + Vue + FastAPI, support DeepAgents、Miner…
README: 语析（Yuxi）是一个基于大模型的智能知识库与知识图谱智能体开发平台。它把 RAG 检索、Milvus 知识库内知识图谱与 LangGraph 多智能体编排整合进统一的多租户工作台：管理员配置知识库、模型与权限，用户在类 ChatGPT 的界面中与可挂载 Skills、MCP、子智能体和沙盒工具的智能体对话，并获得带引用来源、知识图谱推理与可交付产物的…
README: | 前端 | Vue 3 · Vite · Pinia | | 后端 | FastAPI · LangGraph · ARQ (异步 worker) | | 存储 | PostgreSQL · Redis · MinIO · Milvus · Neo4j |
README: **前置要求**：已安装 [Docker](https://docs.docker.com/get-docker/) 与 Docker Compose，并准备至少一个兼容 OpenAI 接口的大模型 API。
README: 等待启动完成后，浏览器打开 `http://localhost:5173`，使用初始化时生成的管理员账户登录即可。

Reviewed at: 2026-07-16T15:52:11Z