📖 The AI Tool Bible

Yuxi

Open-source AI agent platform that fuses agentic RAG with knowledge graphs on a LangGraph runtime.

Free· Free, MIT-licensed self-hostRAGMulti-model
Visit website →
Best for

Pick Yuxi if you're building a self-hosted enterprise assistant and want agentic RAG plus a knowledge graph out of the box.

Skip if

Skip it if you need a managed SaaS, prefer a vector-only RAG stack, or don't have infra capacity to run a LangGraph platform.

Yuxi (语析) is an MIT-licensed AI agent platform built on LangGraph that pairs retrieval-augmented generation with knowledge-graph construction. The runtime gives each agent a sandboxed virtual workspace, supports MCP tool integration, sub-agents for orchestration, and async workers for long-running jobs, plus built-in skills for image generation and report writing.

What sets Yuxi apart is the "agentic RAG" pipeline: it ingests PDFs, Office docs and images, extracts entities and relationships into a knowledge graph, and lets the agent decide when and how to retrieve. It targets teams building internal enterprise assistants who want self-hosted control over their data and aren't satisfied with vanilla vector-only RAG. There is no SaaS tier; you run it yourself.

The platform claims 15+ model providers through a unified config layer (OpenAI, Claude, DeepSeek, and others) and ships connectors for Dify and Notion. Expect the usual self-hosted overhead - infra, model keys, evaluation tuning - in exchange for full ownership.

Editor's take

Yuxi is a credible pick for teams who've outgrown naive vector RAG and want knowledge-graph reasoning without writing their own LangGraph harness. It's clearly aimed at self-hosters and the docs lean Chinese-first, so budget some integration time. Promising, but treat it as a framework, not a finished product.

— The AI Tool Bible editorial team

Pros

  • Open-source under MIT with full self-host control
  • Combines RAG with knowledge graphs rather than vector-only retrieval
  • Sandboxed agent runtime with MCP, sub-agents and async workers
  • Pluggable across 15+ LLM providers via unified config

Cons

  • ⚠️ Self-host only - no managed offering or SLA
  • ⚠️ Smaller community vs. LangChain/Dify; docs lean Chinese-first
  • ⚠️ Knowledge-graph pipeline adds operational complexity over plain RAG

Use cases

agentic-ragknowledge-graphsenterprise-agentsdocument-qamcp-tools

Explore related

Compare with similar tools

All in RAG