Release date: 2026.4.9

Highlights

This release delivers a major end-to-end memory upgrade for UltraRAG, introducing persistent user memory, project memory retrieval, and a dedicated memory-aware RAG demo. It also makes the demo experience significantly more stateful and personalized with SQLite-backed authentication, persistent chat sessions, nickname and model settings management, and knowledge base visibility controls supporting public, private, and shared access.

The frontend is substantially refreshed through a React/Vite-based overhaul, the new Explore section, improved source-detail interactions, and a mobile optimization baseline. We also improve engineering robustness by fixing an async retriever initialization issue, adding compatibility guards for FastMCP 3.x-related builder changes and demo runtime cleanup, introducing CI smoke checks for the SayHello pipeline and frontend build, and adding a structured troubleshooting guide, alongside a broad round of backend and frontend dependency maintenance.

What's Changed

1. Add an end-to-end memory system with persistent user memory, project memory retrieval, and a dedicated memory-aware RAG demo. by @mssssss123 in #379

2. Add SQLite-backed authentication, persistent chat sessions, nickname/model settings management, and knowledge base visibility controls. by @mssssss123 in #379

3. Refactor the frontend with React/Vite, add the Explore section, and improve mobile responsiveness and source-detail interactions. by @mssssss123 in #379

4. Fix async retriever initialization by awaiting retriever_init(...) during startup. by @xhd0728 in #385

5. Fix builder compatibility issues caused by FastMCP 3.x API changes, and harden demo runtime handling around closed-event-loop cleanup noise and backend overrides. by @xhd0728 in #388

6. Add CI smoke checks for the SayHello pipeline and frontend build, then refine the workflow tooling and installation examples. by @xhd0728 in #389 and #390

7. Add a structured troubleshooting guide for UltraRAG workflows. by @onestardao in #352

8. Update README publications and model references. by @xhd0728 in #359

9. Refresh backend and frontend dependencies, including Gradio, vLLM, pypdf, pyjwt, cbor2, cryptography, aiohttp, requests, authlib, orjson, tornado, NLTK, and several frontend packages. by @dependabot in #354, #355, #357, #360, #366, #368, #369, #371, #373, #376, #377, #380, #381, #382, #383, #384, #386, and #387

Release date: 2026.3.10

Highlights

This release introduces expanded document parsing for .doc and .wps formats, alongside new web search capabilities for the demo environment. We have significantly enhanced the chat experience by adding DOCX and markdown export functionality, downloadable text actions, and a streamlined flow for session creation and interruption handling. The citation system has been further optimized with automatic renumbering and improved source sorting for better readability. We also refine the backend performance by introducing concurrency to OpenAI embeddings and add a new Case Study Viewer for extended functionality.

What's Changed

1. Support web search functionality for the demo environment. by @xhd0728 in #285

2. Support .doc and .wps file parsing. by @xhd0728 in #293

3. Implement Case Study Viewer launch functionality. by @mssssss123 in #313

4. Add export functionality for chat messages in DOCX format. by @mssssss123 in #323

5. Simplify the parameter display toggle for a cleaner interface. by @mssssss123 in #282

6. Add user confirmation for interrupting chat generation and streamline the creation of new chat sessions. by @mssssss123 in #305

7. Implement automatic citation renumbering functionality. by @mssssss123 in #314

8. Enhance chat text actions with download capabilities and sort used/unused sources by numeric citation ID. by @mssssss123 in #315

9. Optimize OpenAI embeddings by introducing concurrency. by @xhd0728 in #295

Release date: 2026.1.23

Highlights

Introducing UltraRAG 3.0: Reject "Black Box" Development. Make Every Line of Inference Logic Visible! UltraRAG 3.0 solves the "Last Mile" problem in RAG development, developed by THUNLP, NEUIR, OpenBMB & AI9Stars. This release represents a significant milestone with 129 pull requests merged, bringing major enhancements to functionality, UI/UX, and system stability.

• WYSIWYG Pipeline Builder: From logic to prototype in seconds. Our dual-mode builder (Canvas + Code) syncs in real-time. Click "Build" and your static logic instantly becomes an interactive UI. No more boilerplate code!

• Pixel-Level "White-Box" Visualization: Stop guessing. The "Show Thinking" panel visualizes the entire inference trajectory—loops, branches, and tool calls. Debug bad cases instantly by comparing retrieval chunks vs. model hallucinations.

• Built-in AI Developer Assistant: Stuck on config? The embedded AI Assistant knows the framework inside out. Just use natural language to generate Pipeline configurations, optimize Prompts, or explain parameters.

• DeepResearch Engine: Powered by AgentCPM-Report , it supports "Writing-as-Reasoning." The system dynamically plans, retrieves, and deepens content to generate professional, cited reports automatically.

What's Changed

Demo Mode & Streaming:

• Enable demo-mode compatibility for retriever server and Milvus backend (#145) @mssssss123
• Enable streaming generation in generation server to support demo mode (#147) @mssssss123
• Full stack support (Frontend, Backend, Client) for Demo; Upgrade CLI to v3 (#149) @mssssss123
• Support displaying "thinking process" for demo (#151) @mssssss123
• Support customizable chunk params, optimize corpus server, and add thread cleanup (#167) @mssssss123

Knowledge Base (KB):

• Support knowledge management for demo (#160) @mssssss123
• Support uploading multi-files and folders (#161) @mssssss123
• Support selecting KB during chat (#162) @mssssss123
• Support Chinese KB names (#212) @xhd0728
• Add KB selection validation (#194) @xhd0728
• fix append index error when set an exist collection name (#266 ) @xhd0728

Citations & Rendering:

• Support citation by prompt (#155) @mssssss123
• Support citation for loop-like pipelines (#156) @mssssss123
• Support citation for light research pipelines (#157) @mssssss123
• Re-implement single-step and multi-step citations logic (#173) @mssssss123
• Enhance reference rendering with collapsible unused citations and improved highlight logic (#174) @mssssss123
• Integrate KaTeX for LaTeX rendering in markdown (#171) @mssssss123

Pipeline & Agents:

• Support SurveyCPM pipeline (#183) @mssssss123
• Support background inference suspension & optimize generation details (#185) @mssssss123
• Enhance background task management with user ID isolation (#186) @mssssss123
• Add PaperAgent demo (#189) @mssssss123
• Add AgentCPM-Report demo (#226, #230) @Kaguya-19
• Support Fascmcp v2.14.3 (#218) @xhd0728

AI Assistant:

• Introduce AI Assistant for configuration (#202) @mssssss123
• Enhance AI Assistant capabilities and UI (#203, #229, #231, #233) @xhd0728 @mssssss123

Utilities:

• Add extra_params for generation server (#169) @mssssss123
• Support one-click history cleanup, .docx parsing, and multiline chat input (#172) @mssssss123
• Bundle JS and CSS for offline support (#192) @xhd0728
• Use uv for dependency management (#222) @xhd0728

Logic & Backend:

• Fix server discovery issue when server.py is missing (#152) @mssssss123
• Re-implement search-o1 logic (#158) @mssssss123
• Fix global params bug in client (#184) @mssssss123
• Fix multi-turn chat errors (#193) @xhd0728
• Fix SurveyCPM hard mode parameters and rendering (#205, #208) @Kaguya-19 @mssssss123
• Fix event loop and session expiration issues (#216) @xhd0728
• Fix YAML parsing errors and unicode support (#213, #224) @xhd0728
• Fix image path indexing in generation (#146) @xhd0728

UI & Display:

• Correctly display final step in branch (#170) @mssssss123
• Prevent data misalignment in router mode (#175) @mssssss123
• Fix user role distinction and citation display in multi-chat history (#176) @mssssss123
• Fix log display issues and upgrade MCP version (#178) @mssssss123
• Fix list format checking for global variables (#191) @mssssss123
• Fix AI assistant application error when switching config view (#228) @xhd0728
• Fix text/table rendering and copy function (#237) @xhd0728
• Fix gpu_ids type loading and evaluation result table display (#249) @xhd0728

General:

• Refine chat UI, UX, and support code highlighting (#234) @xhd0728
• Split chat and config views for better workflow (#204) @xhd0728
• Set normal view to chat & enhance session history handling (#220) @xhd0728
• Enhance document preview and UI components (#238) @mssssss123

Layout & Components:

• Redesign knowledge base menu (#168) and management interface (#214) @mssssss123
• Optimize KB dropdown menu & citation display (#206) @xhd0728
• Update homepage layout (#196, #197) @xhd0728
• Update Builder UI components and styles (#232) @mssssss123
• Update navbar functionality and remove status indicator (#221) @mssssss123
• Enhance background task UI (#217) @xhd0728
• Unify alert dialogs (#195) @xhd0728

Code Review & Cleanup:

• Comprehensive code review across multiple servers (Benchmark, Corpus, Custom, Evaluation, Generation, Prompt, Reranker, Retriever, Router, Sayhello) (#239-#248, #250, #251, #255, #257) @xhd0728

Pipeline & Server:

• Refactor retriever server and Milvus implementation for demo (#150) @mssssss123
• Optimize markdown rendering (#153) @xhd0728
• Refactor demo citation and history saving (#154) @mssssss123
• Auto start/stop pipeline execution (#165) @mssssss123
• Optimize reference rendering logic (#166) @mssssss123
• Rename pipelines (#235) @mssssss123
• Remove and update obsolete params/tools in retriever (#200) @mssssss123
• Clear pipelines and unused parameters (#211) @mssssss123

Documentation & Maintenance

• Add environment requirements and usage guide (#190) @mssssss123
• Improve example cases on the welcome page (#180) @mssssss123
• Add V3 tutorial (#236) @mssssss123
• Update README and Docker usage (#227, #252, #254, #256, #261) @mssssss123 @xhd0728
• Security: Add Dependabot configuration and security notice (#258) @xhd0728
• Deps: Bump vllm from 0.13.0 to 0.14.0 (#259) @dependabot

Release date: 2026.1.12

Highlights

This release focuses on improving system stability and core logic. We have addressed a path indexing bug in the generation server to ensure reliable image rendering. Additionally, the search-o1 pipeline has been completely reimplemented, resolving previous implementation flaws.

What's Changed

1. Fixed index reference for image paths in generation server. by @xhd0728 #146
2. Reimplemented the search-o1 pipeline by @mssssss123 @lifelsl #158

Release date: 2025.11.25

Highlights

This release introduces a refreshed UltraRAG front-end UI, fixes several logical issues, and adds new ToolCall and PipelineCall capabilities for directly invoking UltraRAG tools or pipelines from your own code. The retriever server has been further optimized to support full deployment without repeated corpus/index initialization, significantly improving experimental efficiency. We also refine the GPU/CPU configuration logic for more stable and flexible retriever performance.

What's Changed

1. Added full deployment support for the retriever server. by @mssssss123 #136
2. Updated the UltraRAG demo front-end interface. by @mssssss123 #139
3. Optimized GPU/CPU configuration logic for the retriever server. by @mssssss123 #142
4. Introduced new ToolCall and PipelineCall functionality. by @hm1229 @mssssss123 #143

Release date: 2025.11.13

Highlights

This release improves the stability and flexibility of UltraRAG. Retriever and index are now fully decoupled, with added support for Milvus and Faiss. The VisRAG 2.0 pipeline is newly supported, and compatibility with chonkie 1.4.0+ has been updated. Several pipelines, including Search-o1 and WebNote, have been fixed. User experience is improved through progress-bar support during document parsing, a script for saving retrieval results, and zoom-in support for images in the case-study UI. The README now includes a clearer Docker usage guide.

What's Changed

1. Decoupled Retriever and Index with Milvus & Faiss Support: The retriever and vector index are now fully decoupled, with native support for both Milvus and Faiss, providing greater flexibility for large-scale corpus construction and high-performance retrieval. by @mssssss123 #122
2. Added Full Pipeline Support for VisRAG 2.0: UltraRAG now natively supports the VisRAG 2.0 Pipeline, enabling stronger multimodal document understanding and integrated image–text reasoning. by @mssssss123 #115
3. Upgraded Compatibility for chonkie ≥ 1.4.0: Updated document-processing modules to be fully compatible with chonkie 1.4.0+, enhancing stability in structured extraction and parsing workflows. by @yinyannlp @mssssss123 #119 #120
4. Fixed the Search-o1 Pipeline: Resolved critical issues in the iterative search-generation loop, restoring stable functionality for the Search-o1 pipeline. by @mssssss123 #123
5. Fixed Accuracy Metric Calculation Bug: Corrected inaccuracies in the acc metric computation, ensuring more reliable and consistent evaluation results across benchmarks. by @mssssss123 #127
6. Added Script for Saving Retrieval Results: Introduced a new script to persist retrieved content, improving debugging workflows, visualization, and error case analysis. by @mssssss123 #129
7. Fixed Bugs in the WebNote Pipeline: Addressed multiple issues in the WebNote pipeline to improve robustness when processing webpage-like or long-form content. by @mssssss123 #128
8. Enhanced Case Study UI with Image Zoom-In Support: The case-study interface now supports zoomable images, offering a significantly improved multimodal inspection and visualization experience. by @mssssss123 #130
9. Added Progress Bar for Document Parsing: Document parsing now displays a progress bar, providing clearer feedback during long document processing. by @mssssss123 #133
10. Updated README with Complete Docker Usage Guide: The README has been updated with a comprehensive Docker usage section, making deployment easier and more accessible. by @xhd0728 #131

Release date: 2025.10.22

Highlights

1. Comprehensive Multimodal Upgrade: Both the Retriever and Generation Servers now support multimodal inputs, enabling a complete end-to-end multimodal workflow from retrieval to generation.
2. Corpus Parsing and Chunking Redesign: The Corpus Server adds multi-format file parsing with deep MinerU integration, supporting token-level, sentence-level, and customizable chunking strategies to flexibly adapt to diverse corpus structures.
3. Unified Deployment and Efficient Inference: The Retriever and Generation Servers are fully compatible with standardized deployment frameworks such as vLLM, supporting offline inference, multi-engine adaptation, and accelerated experimentation.
4. Enhanced Evaluation and Experimentation Workflow: Introduced TREC-based retrieval evaluation and significance testing modules, supporting parallel experiment execution and multimodal result visualization to optimize research assessment and experimental workflows.

What's Changed

• Corpus Server supports plain text extraction from .txt, .md, .pdf, .xps, .oxps, .epub, .mobi, and .fb2 files. @mssssss123
• Corpus Server adds simple per-page image conversion for .pdf files. @mssssss123
• Corpus Server integrates MinerU for high-precision PDF parsing. @mssssss123
• Corpus Server introduces a new chunking strategy supporting token-level (word/character) segmentation. @mssssss123
• Corpus Server supports sentence-level chunking. @mssssss123
• Corpus Server supports customizable chunking rules (default rule recognizes Markdown sections; other rules can be extended via config files). @mssssss123
• Retriever Server supports three retrieval engines: Infinity, Sentence-Transformers, and OpenAI. @mssssss123
• Retriever Server supports multimodal retrieval. @mssssss123
• Retriever Server adds BM25 sparse retrieval. @xhd0728
• Retriever Server supports hybrid retrieval (dense + sparse). @mssssss123
• Retriever Server provides standardized deployment based on vLLM, unified under the OpenAI-compatible API. @xhd0728
• Retriever Server supports online retrieval via Exa, Tavily, and ZhipuAI. @xhd0728
• Reranker Server supports Infinity, Sentence-Transformers, and OpenAI ranking engines. @xhd0728
• Generation Server supports multimodal inference. @mssssss123
• Generation Server introduces vLLM offline inference, significantly improving experimental efficiency. @mssssss123
• Generation Server supports Hugging Face inference for local debugging. @xhd0728
• Evaluation Server supports TREC retrieval evaluation. @xhd0728
• Evaluation Server supports TREC significance testing. @xhd0728
• VisRAG Pipeline enables an end-to-end workflow from local PDF ingestion to multimodal retrieval and generation. @mssssss123
• RAG Client supports running multiple experiments in parallel under the same pipeline through custom parameter files. @mssssss123
• UltraRAG Benchmark adds six new VQA datasets, including wiki2024 and corresponding VQA corpora. @mssssss123 @xhd0728 @hm1229
• Case Study UI adds multimodal result visualization support. @mssssss123

Release date: 2025.08.28

Highlights

UltraRAG 2.0 introduces a complete redesign of the system architecture, fully adopting the MCP-based modular structure. This release significantly improves flexibility, extensibility, and developer experience, enabling researchers and engineers to rapidly build, customize, and reproduce complex RAG pipelines with minimal overhead.

The core of UltraRAG 2.0 is the new MCP Server–Client workflow, providing a clean separation between functional components such as Corpus, Retriever, Generation, Evaluation, Router, and Note. The pipeline engine has been rebuilt to support looping, branching, multi-server instances, and fully customizable data flow, enabling Search-o1, RankCoT, WebNote, Iter-RetGen, and other advanced RAG pipelines to run out-of-the-box.

Developer experience is further enhanced with streamlined YAML configuration, Functool-style tool invocation, automatic server building, and improved error handling. UltraRAG 2.0 marks a major step toward a unified, reproducible, and developer-friendly RAG research platform.

What’s Changed

MCP Architecture & Server–Client Redesign

• Unified MCP Server Architecture: All RAG components (Benchmark, Corpus, Retriever, Generation, Evaluation, Router, Note) are now implemented as standalone MCP Servers, enabling modular development and clean extensibility.
  by @hm1229 @xhd0728
• Full MCP Client Pipeline Engine: Introduced a new pipeline executor supporting loop, branch, router, multi-server aliases, early stopping, and nested pipelines, making complex iterative workflows natively supported.
  by @hm1229 @xhd0728

Corpus Server

• Added unified file-reading support (doc/docx/pdf/md/txt) and multiple chunking strategies (static/semantic/LLM).
  by @mssssss123
• Enhanced Chunk Pipeline: Improved stability and extensibility for large-document processing and JSONL corpus generation.
  by @mssssss123

Retriever & Reranker Server

• Retriever Server Rebuilt: Embedding, indexing, retrieval, and reranking fully modularized. by @mssssss123 @Kaguya-19 @gdw439
• Improved Vector Index Handling: Cleaner index-building API and improved fault tolerance during large-scale indexing.
  by @mssssss123 @Kaguya-19 @gdw439
• Reranker Server Added: Native support for passage reranking in modular form. by @mssssss123 @xhd0728

Prompt & Generation Server

• Unified Generation Server: Supports OpenAI-compatible API, local HF/vLLM models, customizable sampling params, and Prompt Server integration. by @methanechen @hm1229
• Prompt Server Support: Introduced dedicated Prompt Server for parameterized Jinja templates and reusable prompt logic.
  by @hm1229 @xhd0728

Benchmark & Evaluation Server

• Unified Benchmark Loader: Added support for multiple datasets simultaneously with standardized q/gt extraction.
  by @lifelsl @xhd0728
• Expanded Metrics: Added CoverEM, Acc, and improved EM/F1 calculation consistency.
  by @xhd0728 @lifelsl

Baseline Pipeline Implementations

• baselines added and tested on UltraRAG 2.0 MCP.
  by @xhd0728 @Mignonmy @lifelsl
• Multi-Benchmark Leaderboard Support: Standardized evaluation outputs to enable stable leaderboard reproduction.
  by @xhd0728 @mssssss123

Developer Experience Improvements

• ultrarag build Command: Automatically compiles servers, generates server.yaml & parameter.yaml, and validates parameters.
  by @hm1229
• Improved Logging & Debugging: Added log-level control, error propagation, and clearer exception tracing.
  by @hm1229
• Multi-Agent/Multi-Retriever Support: Added server aliasing for running multiple retrievers or generators within one pipeline.
  by @xhd0728

Documents

• Project Website: The official UltraRAG 2.0 website. by @mssssss123 @xhd0728
• Comprehensive Tutorial: A full end-to-end guide covering installation, MCP Server definitions, pipeline configuration, loop & branch logic, example baselines, and best practices for developing new servers and tools. by @mssssss123 @xhd0728
• README: The README has been redesigned to include a clearer introduction to UltraRAG 2.0. by @mssssss123 @xhd0728

Contributors

Thanks to everyone who contributed to the UltraRAG 2.0 beta and release:

• Project Leaders: @yanyk13 @mssssss123
• Core Contributors: @xhd0728 @hm1229 @mssssss123
• Contributors: @lifelsl @methanechen @Kaguya-19 @gdw439 @Mignonmy