The official documentation for UltraRAG can be found at ultrarag.openbmb.cn. This website serves as the primary resource for detailed information, tutorials, and guides related to the UltraRAG framework. Additionally, the project maintains a comprehensive GitHub repository at github.com/OpenBMB/UltraRAG, which also includes documentation within its docs folder, including a Chinese version README_zh.md. These resources are developed and maintained by the collaborative efforts of Tsinghua University’s THUNLP lab, Northeastern University’s NEUIR lab, OpenBMB, and AI9stars.
The UltraRAG documentation provides extensive guidance for both researchers and developers looking to utilize this low-code, modular RAG framework. It covers various aspects, from getting started with basic RAG pipelines to deploying complex demo systems and conducting scientific experiments. The documentation is structured to help users understand how to write pipeline configuration files using YAML, compile and adjust parameters, and run pipelines. It also includes information on the framework’s core architecture, which standardizes RAG components like retrievers, generators, and evaluators as independent Model Context Protocol (MCP) servers.
For developers, the documentation offers insights into integrating various backend technologies, including support for local models via vLLM or Hugging Face Transformers, as well as API-based models. It also provides deployment guides that cover the setup of retrievers, generation models, and vector databases like Milvus. The documentation emphasizes UltraRAG’s features, such as its low-code visual orchestration with bidirectional synchronization between canvas and code, modular MCP servers for enhanced reusability, and built-in evaluation suites for benchmark comparison. The official site also provides quick start guides, installation instructions, and detailed explanations of how to leverage UltraRAG’s capabilities for developing and evaluating advanced RAG systems.