update 3.0 blog (#263)

* update 3.0 pr blog

* Refine content and remove placeholder references

Removed placeholder references and improved clarity in the document.

* Revise UltraRAG 3.0 documentation for clarity

Updated content to enhance clarity and structure of UltraRAG 3.0 features, including interactive development and debugging capabilities.

* Remove introductory paragraph on white-box debugging

Removed introductory paragraph about showcasing scenarios.

Author: mssssss123

blog/cn/ultrarag3_0.md

@@ -0,0 +1,57 @@
+# UltraRAG 3.0 发布：拒绝“盲盒”开发，让每一行推理逻辑都清晰可见
+
+**“验证算法原型只需一周，构建可用系统却耗时数月。”** 这句看似调侃的“吐槽”，却是每一位算法工程师不得不面对的真实困境。
+
+今天，清华大学 THUNLP 实验室、东北大学 NEUIR 实验室、OpenBMB 、面壁智能与 AI9Stars 联合发布 **UltraRAG 3.0** ，针对上述痛点，为科研工作者与开发者打造更懂开发者的技术框架，具备 3 大核心优势：
+
+- **从逻辑到原型的一键跨越，让算法工程师回归“算法”：** 提供“所见即所得”的 Pipeline 构建器，自动接管繁琐的界面封装。只需专注于逻辑编排，即可让静态代码即刻变身为可交互的演示系统。
+- **全链路白盒化，推理轨迹的“像素级”可视化：** 打造“透明化”的推理验证窗口，实时呈现模型在复杂长链条任务中的每一次循环、分支与决策细节。
+- **内置智能开发助手，你的“交互式开发指南”：** 内嵌懂框架的 AI 助手，通过自然语言交互辅助生成 Pipeline 配置与优化 Prompt，大幅降低上手门槛。
+
+
+
+## 逻辑即应用从编排到交互的「零距离」体验
+
+让算法的终点不再是冷冰冰的控制台日志。UltraRAG 3.0 通过自动化处理繁琐的界面封装与参数对接，确保在逻辑编排完成的一刻，便已同步生成了可交互的演示界面：
+
+- **配置即应用：** 只需定义 Pipeline 的 YAML 配置文件，框架即可将其自动解析并转化为标准的交互式 Demo。
+- **双模式构建器：** 为了平衡易用性与灵活性，我们打造了可视化与代码实时同步的构建引擎：
+  - 画布模式：通过 UI 组件，像搭积木一样直观组装 Loop（循环）、Branch（分支）等复杂逻辑。
+  - 代码模式：直接编辑 YAML 配置文件，画布视图实时渲染更新，满足开发者对参数微调的精准控制需求。
+- **一键编译验证：** 构建完成后，点击顶部的 “Build” 按钮，系统自动执行逻辑自检与语法校验，并动态生成参数配置面板。参数就绪的瞬间，静态的算法逻辑即刻变身可交互的系统，真正实现“所写即所得，所得即所用”。
+
+## 拒绝「黑盒」让复杂 RAG 的推理脉络清晰可见
+
+随着 RAG 技术从简单的单次检索向多轮动态决策演进，推理链路往往长达上百个 Step。在缺乏中间态监控的情况下，调试过程就如同在迷雾中从头再来，错误定位全靠“猜”。
+
+UltraRAG 3.0 重新定义了 **Chat 界面**——它不仅是用户交互的入口，更是 **逻辑验证的窗口**。我们深知，对于开发者而言，知道“结果是什么”远远不够，看清“结果怎么来的”才是优化的关键。
+
+我们通过“Show Thinking”面板，对系统“思考”的全过程进行了像素级的实时可视化——从复杂的循环分支到具体的工具调用，**所有中间状态均以结构化的形式流式呈现**。即使是 DeepResearch 这样复杂的长流程任务，开发者也能实时掌握执行进度，让过程不再是漆黑的等待。当 Bad Case 出现时，开发者无需再后台翻找日志，只需在界面上直接比对检索切片与最终答案，快速判断问题是出在“数据层的噪声”还是“模型层的幻觉”，极大地缩短了优化迭代的周期。
+
+
+## 解放定制开发的「框架桎梏」
+
+想要尝试一个新的算法逻辑，往往需要深入框架底层，重写大量继承类。为了实现 **10%** 的核心算法创新，却不得不背负 **90%** 的框架学习成本。
+
+**UltraRAG 3.0 将整套开发文档与最佳实践内嵌到了框架自带的智能助手中**。它或许无法像 Cursor 那样帮你写完整个项目，但它绝对是最懂 UltraRAG 的高效辅助工具。通过自然语言交互，它能帮你彻底抹平“阅读文档”与“编写配置”之间的认知阻力：
+
+- **配置生成：** 你只管描述需求（例如：“我要一个带多路召回和重排序的流程”），助手即可自动生成标准的 Pipeline 结构草稿，只需微调即可直接使用。
+- **Prompt 调优：** 助手能根据当前任务上下文，提供针对性的 Prompt 优化建议，快速适配特定业务场景。
+- **辅助理解：** 看不懂某个参数或逻辑？无需跳转浏览器翻阅官网文档，直接提问即可获得开发建议与代码示例，让编码过程不再中断。
+
+### 实战演示：它能帮你做什么？
+
+我们在此展示了四个真实的交互场景，看看它是如何将自然语言转化为“可执行逻辑”的：
+
+1. 结构调整：一句话修改 Pipeline
+> User：“请帮我修改当前的 Pipeline，增加一个引用（Citation）功能模块，以便对生成内容进行事实核查。”
+
+2. 场景适配：针对性优化 Prompt
+> User: “我需要针对 法律场景 优化当前的 Prompt。请调整提示词，使其生成的回答在该领域的术语使用和逻辑推演上更加专业和准确。”
+
+3. 配置调整：轻松修改底层参数
+> User: “我要切换生成后端的配置。请将生成模型后端改为 OpenAI，模型名更改为 qwen3-32b，API 服务部署在端口 65503 上。”
+
+4. 自由调优：从概念到实现的捷径
+>User: “我想参考这篇论文：https://arxiv.org/pdf/2410.08821 (DeepNote)，来重新设计我的 RAG 流程。请分析文章中的核心思想，并帮我搭建一套类似的 Pipeline 架构。”
+   

blog/en/ultrarag3_0.md

@@ -0,0 +1,65 @@
+# UltraRAG 3.0 Release: No More “Blind-Box” Development — Every Line of Reasoning Made Explicit
+
+**“Validating an algorithmic prototype takes only one week, but building a usable system can take months.”**  
+This seemingly tongue-in-cheek remark captures a very real pain point faced by almost every algorithm engineer.
+
+Today, **UltraRAG 3.0** is jointly released by **THUNLP Lab at Tsinghua University, NEUIR Lab at Northeastern University, OpenBMB, ModelBest, and AI9Stars**. Designed specifically to address this challenge, UltraRAG 3.0 delivers a developer-centric framework for both researchers and practitioners, built around three core advantages:
+
+- **From logic to prototype in one step — letting algorithm engineers focus on algorithms:**  
+  A true *what-you-see-is-what-you-get* Pipeline Builder automatically handles tedious UI integration. By focusing solely on logical orchestration, static code is instantly transformed into an interactive demo system.
+- **End-to-end white-box transparency with pixel-level visualization of reasoning traces:**  
+  A transparent reasoning inspection interface exposes every loop, branch, and decision made by the model in complex, long-horizon tasks.
+- **An embedded intelligent development assistant — your interactive development guide:**  
+  A framework-aware AI assistant supports Pipeline generation and prompt optimization through natural language interaction, dramatically lowering the barrier to entry.
+
+## Logic Becomes Application: Zero Distance from Orchestration to Interaction
+
+Let algorithms move beyond cold console logs. By automatically handling UI encapsulation and parameter binding, UltraRAG 3.0 ensures that the moment logical orchestration is complete, a fully interactive demo interface is generated simultaneously.
+
+- **Configuration as Application:**  
+  Simply define a Pipeline using a YAML configuration file, and the framework automatically parses it into a standardized interactive demo.
+- **Dual-Mode Builder:**  
+  To balance usability and flexibility, we provide a construction engine in which visualization and code remain fully synchronized:
+  - **Canvas Mode:** Assemble complex logic such as Loops and Branches intuitively through UI components, like building with blocks.
+  - **Code Mode:** Directly edit the YAML configuration while the canvas view updates in real time, enabling precise parameter-level control.
+- **One-Click Build and Validation:**  
+  After construction, clicking the “Build” button triggers automatic logic checks and syntax validation, followed by dynamic generation of parameter panels. Once parameters are set, static algorithmic logic instantly becomes a fully interactive system — truly achieving *what you write is what you get, and what you get is immediately usable*.
+
+## Rejecting the “Black Box”: Making Complex RAG Reasoning Fully Transparent
+
+As RAG systems evolve from single-pass retrieval to multi-round dynamic decision-making, reasoning chains can easily span hundreds of steps. Without visibility into intermediate states, debugging becomes a process of blind trial and error.
+
+UltraRAG 3.0 redefines the **Chat interface** — it is not only the entry point for user interaction, but also a **window for logic validation**. For developers, knowing *what* the result is matters far less than understanding *how* it was produced.
+
+Through the “Show Thinking” panel, UltraRAG provides pixel-level, real-time visualization of the entire reasoning process. From complex loops and branches to concrete tool invocations, **all intermediate states are streamed and presented in a structured form**. Even for long-horizon workflows such as DeepResearch, developers can monitor execution progress in real time, eliminating opaque waiting periods.
+
+When bad cases arise, there is no need to search through backend logs. Developers can directly compare retrieved evidence slices with the final answer in the interface to quickly determine whether the issue originates from **data-level noise** or **model-level hallucination**, significantly shortening the iteration cycle.
+
+## Breaking Free from Framework Lock-In
+
+Experimenting with new algorithmic ideas often requires diving deep into framework internals and rewriting large numbers of inherited classes. To realize **10%** of genuine algorithmic innovation, developers are frequently forced to bear **90%** of the framework learning cost.
+
+**UltraRAG 3.0 embeds its complete documentation and best practices directly into an intelligent, framework-aware assistant.** While it may not write an entire project for you like Cursor, it is purpose-built to deeply understand UltraRAG. Through natural language interaction, it eliminates the cognitive gap between “reading documentation” and “writing configuration”:
+
+- **Configuration Generation:**  
+  Simply describe your requirement (e.g., “I want a pipeline with multi-path retrieval and reranking”), and the assistant generates a standard Pipeline structure draft that can be used with minimal modification.
+- **Prompt Optimization:**  
+  Based on the current task context, the assistant provides targeted prompt optimization suggestions to rapidly adapt to specific application scenarios.
+- **Assisted Understanding:**  
+  Confused by a parameter or logic component? No need to leave your editor to search the documentation. Ask directly and receive development guidance and code examples without interrupting your workflow.
+
+### Practical Demonstrations: What Can It Do for You?
+
+Below are four real interaction scenarios illustrating how natural language is translated into **executable logic**:
+
+1. **Structural Adjustment: Modify a Pipeline in One Sentence**  
+   > User: “Please modify the current Pipeline by adding a citation module for factual verification of generated content.”
+
+2. **Scenario Adaptation: Domain-Specific Prompt Optimization**  
+   > User: “I need to optimize the current Prompt for the legal domain. Please adjust it so that the generated responses use more professional terminology and more rigorous logical reasoning.”
+
+3. **Configuration Update: Easily Switch Backend Parameters**  
+   > User: “I want to switch the generation backend. Please change the generation model backend to OpenAI, set the model name to qwen3-32b, and deploy the API service on port 65503.”
+
+4. **Free-Form Optimization: From Concept to Implementation**  
+   > User: “I want to redesign my RAG workflow by referencing this paper: https://arxiv.org/pdf/2410.08821 (DeepNote). Please analyze its core ideas and help me construct a similar Pipeline architecture.”