rag

介绍 ASK，一款基于本地 LLM 和检索增强生成（RAG）的独立桌面知识助手应用，支持 Qwen、Llama 等多种本地模型，使用 Python 与 CustomTkinter 构建。

LlamaIndex 发布 Document Parsing Protocol (DPP) 工作草案，为文本、表格、表单、图表、化学结构和乐谱等文档解析决策提供统一规则集，包含 140 条规则并提供 Claude 技能与 ChatGPT 参考文件供试用反馈。

该论文研究在RAG知识冲突分类任务中，使用typed decision models替代大语言模型的可行性，结果表明决策模型在4-39倍低成本和更低延迟下可匹配或超越Gemini 2.5 Flash的准确率。

A technical report compares taxonomy-label search with a two-stage embedding retrieval plus LLM filtering approach on a corpus of 7,354 randomized trials, finding the new method recovers 92% of systematic-review evidence versus 45% for label search, at a cost that scales with search depth.

RouteDRAG 是一个基于学习的框架，用于在多模态文档 RAG 中决定是否调用 Agent 及其执行路径，在两个基准上比仅用短路径执行平均提升 3.47 个百分点，且仅将 79.3% 的查询分配给 Agent。

该论文提出 LM-DSB，一个结合 RAG 与本地部署 LLM 的轻量级多智能体系统，用于自动生成建筑工地每日安全简报，在消费级硬件上实现了接近人工质量的文档生成与专家认可。

OA-RAG is an agentic framework that combines formal ontology-based reasoning with LLM reasoning to improve faithfulness in retrieval-augmented generation, reporting grounding levels (entailed, supported, related, unrelated) and achieving top performance on medical QA benchmarks derived from large-scale ontologies.

Memvid is an open-source AI memory layer tool that packages raw data, embeddings, BM25 and HNSW search indexes, and metadata into a single .mv2 file. It supports offline operation, version management, and failure recovery, and aims to replace complex RAG pipelines and vector databases. It claims state-of-the-art performance in long-term conversational memory on the LoCoMo benchmark, with extremely low retrieval latency.

A recommendation for Stanford's CS329Z 'Engineering AI Agents' course, which systematically covers the core of agent engineering: LLM fundamentals, RAG, tool calling and function calling, framework orchestration (DSPy, LangGraph), evaluation and reliability, and more — signaling that agent development is shifting from an API-patching craft into a serious engineering discipline.

Survival LLM is a free, open-source AI assistant and emergency reference library that runs fully offline on your own computer, answering first-aid and disaster-preparedness questions with citations from trusted sources like FEMA, CDC, and Wikipedia.

一篇科普长文，借 OpenAI 分阶段关停自助微调平台（2027 年 1 月起全面停止新微调任务）这一新闻切入，通俗解释模型预训练、微调与 RAG 三者的区别，指出微调适合调整风格而非注入业务知识，而真正掌握企业数据的是 RAG。

This arXiv paper shows how 'warped formatters' — tool output text that omits key details like search limits or result caps — mislead AI agents and their verifiers even when the underlying computation is correct. Repaired formatters drastically cut unsupported claims and flipped verifier acceptance from 90/90 wrong to 90/90 correct, and an audit prompt rediscovered live defects in the authors' own system plus OpenClaw and Hermes, prompting merged fixes.

SOPO（Socratic Guided Policy Optimization）提出了一种多轮强化学习框架，通过苏格拉底式教师引导反馈和基于轮次的优势缩放来解决span级幻觉检测中的奖励稀疏与长度膨胀问题，在RAGTruth上取得SOTA，超越GPT-5等模型并开源了1.7B和4B模型。

The paper introduces a framework of 'memory benefit' and 'memory tax' to measure the net value of dedicated agent memory systems over naive RAG, finding that many memory systems underperform direct retrieval across benchmarks. It analyzes how memory processes can remove or obscure evidence and calls for evaluating the net value of memory rather than standalone performance.

该论文提出了一套面向离线边缘生成式AI的连通性体制框架（C0-C4）与推理时信任机制分类法，并在Raspberry Pi 5上以Qwen2.5模型进行本地RAG实证研究，发现签名知识包可被投毒攻击绕过，且现有验证机制无法保证证据真实性。

A free online book, 'LangGraph in Small Pieces,' teaches building AI agents from fundamentals to production using real local-model runs, hands-on challenges, and a public GitHub repo. It covers state, nodes, tool calling, MCP, streaming, RAG, observability, and deploys, pinned to LangGraph 1.2.14.

A practitioner ran a four-agent RAG panel experiment and documented three unexpected failure modes: irrelevant-but-supportive citations, agents citing their own generated notes, and fixed top-K retrieval forcing filler citations. The post shares fixes in progress and asks the community how they measure real RAG improvement.

A developer is asking the community how they actually solve agent memory in production, whether through vector DBs with RAG, knowledge graphs, or human-in-the-loop approval for a shared knowledge base.

作者探讨构建研究型 AI 智能体时的两种架构选择：直接调用一体化 deep research API（Gemini Deep Research、OpenAI、Tavily Research 等），或基于搜索/抓取原语（Firecrawl、Exa、Tavily）自建规划与合成循环，并就质量、可观测性、扩展性与成本估算向生产环境实践者提问。

论文提出开源 RAG 受控评测框架 LASAGNA，在八个基准和 5K-200K 文档语料上测试 450 种配置，发现单独有效的 RAG 技术组合后收益大幅缩水（预期 +47% 仅得 +17%），检索质量与答案忠实度的设计取舍互相冲突，且最优组合随语料规模变化。