UltraRAG Releases

This release focuses on security patches, performance enhancements, and architectural refactoring. It addresses CVE vulnerabilities in Starlette, introduces parallel processing for community labeling, and modularizes the extraction logic.

Jun 25, 2026

Introduces a self-improving 'work memory' system that records outcomes of Q&A sessions to generate a `LESSONS.md` file for future sessions. It also fixes extraction bugs regarding qualified class-method calls and non-hashable nodes.

Jun 24, 2026

Significant performance improvements via trigram query prefiltering and the addition of CUDA support for C++ code extraction. It also resolves phantom duplicate nodes and enhances the skill runbook.

Jun 23, 2026

Enhances path portability to ensure artifacts work across clones and machines. It fixes hyperedge generation for native backends and ensures the `GRAPHIFY_OUT` environment variable is respected throughout the toolchain.

Jun 22, 2026

Fixes critical crashes and data loss issues in the skill-runbook generation, particularly regarding semantic chunking and zero-node guards. It also resolves FileSlice handling issues in the semantic extractor.

Jun 19, 2026

This major release delivers an end-to-end memory system for UltraRAG with persistent user memory, project memory retrieval, and a dedicated memory-aware RAG demo. It also introduces SQLite-backed authentication, persistent chat sessions, nickname/model settings management, and knowledge base visibility controls. The frontend is substantially refreshed with React/Vite, a new Explore section, and mobile optimization.

Apr 9, 2026

This release introduces expanded document parsing for .doc and .wps formats alongside web search capabilities for the demo environment. The chat experience is enhanced with DOCX and markdown export functionality, downloadable text actions, and improved citation system with automatic renumbering and source sorting.

Mar 10, 2026

UltraRAG 3.0 introduces a major paradigm shift toward transparent RAG development with 129 pull requests merged. This release features a WYSIWYG Pipeline Builder, pixel-level white-box visualization for debugging inference trajectories, an embedded AI Developer Assistant for natural language pipeline configuration, and a DeepResearch Engine powered by AgentCPM-Report for automated report generation.

Jan 26, 2026

This release focuses on stability improvements and core logic fixes, addressing a critical path indexing bug in the generation server and completely reimplementing the search-o1 pipeline to resolve previous implementation flaws.

Jan 12, 2026

This release introduces a refreshed UltraRAG front-end UI, new ToolCall and PipelineCall APIs for programmatic tool/pipeline invocation, and optimized retriever server deployment with improved GPU/CPU configuration logic.

Nov 25, 2025

This release improves system stability with decoupled retriever and index architecture supporting Milvus and Faiss, adds VisRAG 2.0 pipeline support, and enhances the case study UI with image zoom capabilities and progress feedback during document parsing.

Nov 13, 2025

Nov 9, 2025

This major release provides a comprehensive multimodal upgrade enabling end-to-end multimodal workflows, redesigns corpus parsing with deep MinerU integration and flexible chunking strategies, and adds unified vLLM-based deployment with enhanced evaluation capabilities including TREC evaluation and significance testing.

Oct 22, 2025

A major version release introducing a complete MCP-based architecture redesign. The system is now modularized into distinct servers for various RAG components and features a new pipeline engine supporting complex workflows.

Oct 21, 2025

This release adds support for loading configuration rules via protocol-based files for various proxy parameters, offering flexibility through filename, base64, or direct string formats.

Sep 2, 2025

Jun 11, 2025

May 11, 2025

May 11, 2025

Apr 1, 2025

This release was originally intended to be version 1.0.0 but was downgraded due to stability concerns. It introduces configuration file support, custom subprocess adapters, and a complete migration to an asynchronous codebase.

Jan 16, 2024