实时增长榜 · 周综述每周日由 spark 生成
潜入 Shell 的幽灵。Ante 是一个自包含的 Agent 框架，核心高度优化。体验类似 Claude Code 或 Codex，但无其依赖与模型限制。Ghost in your shell. Ante is a self-contained agent harness with a highly optimized core. It works like Claude Code or Codex, with none of their dependencies or model constraints.
herdr 的原生 macOS 控制台，集中管理所有编码 agents 及其实时终端，并支持跨设备使用Native macOS console for herdr — all your coding agents and their live terminals, across devices
面向你的 AI Agent 的开源控制平面。连接工具、雇佣 Agent、追踪每一个 token 与每一分钱。Open-source control plane for your AI agents. Connect tools, hire agents, track every token and dollar
将任意网站转换为面向 AI agents 的精简 CLI，仅用数百 token 而非数万个 token 即可浏览网页Turn any website into a compact CLI tailored for AI agents. Browse the web in hundreds of tokens, not tens of thousands.
原生 iOS 应用数据浏览器、便携补丁管理器、简易清理工具，以及 PosterBoard 壁纸实验室。Native iOS app-data browser, portable patch manager, limited cleaner, and PosterBoard wallpaper lab.
面向组织级 Agent 的开源控制平面与运行时：共享公司上下文、隔离执行、审批流与 MCP。Open-source control plane and runtime for organisational agents: shared company context, isolated execution, approvals and MCP.
🔥🔥🔥 企业级 AI 中间件，可替代 unifyapps、n8n、lyzr🔥🔥🔥 Enterprise AI middleware, alternative to unifyapps, n8n, lyzr
Qwen3.8 27B 在 SGLang 上运行于 DGX SparkQwen3.8 27B on SGLang for DGX Spark
把本机 Wallpaper Engine 的壁纸变成 DSH 网页界面的背景：Video 动态播放、Web 以 iframe 加载、Scene 壁纸提取主纹理作为静态帧；iOS 液态玻璃设置窗口（配色 / 玻璃颜色 / 透明度）、内容分级与类型过滤、自定义壁纸上传、紧凑 CD 架布局、黑胶唱片展示、隐藏 / 恢复、倍速 / 翻转与自动轮播。感谢 Jerry 维护 macOS 版。
DeepSeek Harness 的开源 Windows 桌面客户端与 GUI —— 零配置安装包，内置 Codex、plugins、skills、SSH、移动端远程访问与 11 套皮肤。Open-source Windows desktop client and GUI for DeepSeek Harness — zero-setup installer with Codex, plugins, skills, SSH, mobile remote access, and 11 skins.
DeepSeek Harness plugin store, marketplace and hub — 3,100+ dsh plugins with search, rankings, install commands and a free public API. DeepSeek Harness 插件市场 / 插件商店：自动收集与格式校验，免费搜索 API。deepseek1024.com
最全面的DeepSeek Harness🐋启动器 *首创多实例管理中枢* | 多开并行 | 整合包下载 | 一键部署 | 插件管理 |
Pothole Reporter 是一款独立的 Android 应用，用于记录坑洼、垃圾与未加盖窨井，支持后台 Drive Mode、马哈拉施特拉邦范围的工单交接以及已审核的巡查路线，覆盖已支持的印度地区。Pothole Reporter is an independent Android app for documenting potholes, garbage and open manholes, with background Drive Mode, Maharashtra-wide handoffs and reviewed routes across supported Indian areas.
DeepSeek Harness 安卓启动器：内置 proot+Ubuntu，免 ROOT 免 Termux 一键运行 deepseek-harness，多源测速/分步安装/WebUI 预览/设备 Shell 桥接
面向法律与重型文档工作流的 AI-native IDE 工作空间：文件、Agent、插件、WPS 编辑、OCR、证据链。律师的 VS Code。AI-native IDE workspace for legal and document-heavy workflows: files, agents, plugins, WPS editing, OCR, evidence chains. VS Code for lawyers.
通过扫码或机器人凭据把IM机器人接入DeepSeek Harness（支持飞书、微信、钉钉、企业微信、QQ、Slack、Telegram、Discord和WhatsApp）。 Connect IM bots to DeepSeek Harness via QR code or credentials (9 channels).
DeepSeek Harness 的可视化非线性对话工作区插件；DeepSeek Harness 基于画布的会话浏览器与分支工作区。A visual, non-linear conversation workspace plugin for DeepSeek Harness ; A canvas-based session explorer and branching workspace for DeepSeek Harness.
RuvNet Brain——基于 Reuven Cohen（rUv）的 RuvNet 技术栈（RuVector/RVF、Ruflo、AgentDB、RuLake、SPARC 及 21 个构建块）为 Claude Code 提供一个可下载、源码溯源的"大脑"。通过单一 MCP 工具（search_ruvnet）让 Claude 立足真实源码，而不是脱离技术栈空想。RuvNet Brain — a downloadable, source-grounded brain for Claude Code over Reuven Cohen's (rUv's) RuvNet stack: RuVector/RVF, Ruflo, AgentDB, RuLake, SPARC + 21 building blocks. Grounds Claude in real source via one MCP tool (search_ruvnet), so it builds with the stack instead of drifting off it.
AI 编程智能体统一桌面工作台 | Desktop hub for managing Claude Code, Codex CLI, Gemini CLI & OpenCode. Tauri 2 · Rust · React · SQLite.
为你的 AI 助手提供真实的网络搜索、整页阅读与多源研究能力，引用内容绝不伪造。Give your AI assistant real web search, full-page reading, and multi-source research with citations that are never fabricated.
将 Jira 与 Confluence 镜像到单个本地 SQLite 文件——macOS 应用、Web UI 与 Agent SQL/MCP。即时搜索、离线可用，单一二进制。Your Jira and Confluence, mirrored into one local SQLite file — macOS app, web UI, and agent SQL/MCP. Instant search, offline, one binary.
面向行情录制、交易安全与可复现研究的工程工具包。多交易所 kill-switch、带钱包标签的 Hyperliquid 微结构录制器、资金费率 carry 研究栈。Engineering toolkit for market-data recording, execution safety and reproducible research. Multi-exchange kill-switch, wallet-tagged Hyperliquid microstructure recorder, funding-carry research stack.
为 AI Agent 准备的考试评分知识脑：粘贴文档 URL，即可获得基于 MCP 的可搜索知识脑，并附带测量分数与已知缺口。AGPL。Exam-scored knowledge brains for AI agents: paste a docs URL, get a searchable brain over MCP with a measured score and known gaps. AGPL.
Overleaf 的替代品，在 LaTeX 代码编辑器内置类 Antigravity 的 agentic AI，每个项目免费支持不限数量的协作者。An Overleaf alternative with an inbuilt agentic AI (like Antigravity) in LaTeX code editor..with unlimited collabrators for each project for free
面向 LLM Agent 的本地优先上下文引擎——在检索路径上无需 LLM 调用即可检索、权衡并压缩代码库到上下文窗口。采用 SQLite 知识存储，支持 CLI + HTTP + MCP，提供 know/miss Agent 契约。Local-first context engine for LLM agents — retrieves, weighs, and compresses your codebase into the context window with no LLM call on the retrieval path. SQLite knowledge store; CLI + HTTP + MCP; know/miss agent contract.
在工作发生的地方就地捕获团队知识，由 agent 归档，并以可核验的引用形式给出回答。A team's knowledge, captured where the work happens, filed by an agent, and answered with citations you can check.
针对 DeepSeek Harness Agent 的策略强制、证据优先的治理工作流。Policy-enforced, evidence-first governed workflows for DeepSeek Harness agents.
原生 LaTeX 的修订追踪，面向人类与 AI 辅助审阅。LaTeX-native track changes for human and AI-assisted review
面向论文、学位论文、文献综述与稿件的开放学术写作短语库、修辞策略与循证安全的研究写作技能。Open academic and scientific writing phrasebank, rhetorical moves, and evidence-safe research writing skills for papers, theses, literature reviews, and manuscripts.
用于实时技术分析与市场执行的自主 agentic 框架。Autonomous agentic framework for real-time technical analysis and market execution.
实验性 Python 库：基于 dataframe 导航至研究问题，并内置统计与因果安全闸门Experimental Python library for dataframe-to-research-question navigation with statistical and causal safety gates.
Hyper Terminal 2026：下一代 Web 技术 CLI ⚡ | 免费开源Hyper Terminal 2026: Next-Gen Web Tech CLI ⚡ | Free & Open Source
本地优先的 Markdown、Typst 与 LaTeX 编辑器，支持桌面端与移动端。A local-first Markdown, Typst and LaTeX editor for the desktop and mobile
LLM 的工作记忆。一个供 agent 钉存与更新的会话图，避免将所有内容倾倒到聊天中。Working memory for LLMs. One session graph that agents pin and update — without dumping everything into chat.
非官方的 MCP server，用于 AI 驱动的 NI Multisim 电路生成、仿真、数据导出与报告。Unofficial MCP server for AI-driven NI Multisim circuit generation, simulation, data export, and reports
TickTick CLI 精通 2026 —— 终极任务管理器、MCP 与 Skill 中心。TickTick CLI Mastery 2026 - Ultimate Task Manager, MCP & Skill Hub
面向 AI Agent 工具调用的确定性 <0.1ms 不变量清算内核。零外泄 AST SQL 防火墙、数值速率限制与 Merkle GRC 审计追溯。Deterministic <0.1ms Invariant Clearance Kernel for AI Agent Tool Calls. Zero-egress AST SQL firewall, numeric velocity limits & Merkle GRC audit trails.
面向业务的 RAG | 要么有引用，要么不要 | 客服团队需要一份可核验的答案RAG for business | Citations or nothing | Support teams need an answer they can verify
Agent Fridge——阻止 AI 编码 agent 相互覆盖彼此的成果。提供路径声明、租约、一次写入笔记，并为任意终端生成共享看板。Agent Fridge - stop AI coding agents from overwriting each other's work. Path claims, leases, write-once notes, and a generated shared board for any terminal.
Chat2DB Community 的 Rust 产品运行时，配套 Java 数据库兼容引擎Rust product runtime with Java database compatibility engine for Chat2DB Community
AI 教材数字化与互动辅导平台：PDF/扫描教材经 MinerU OCR 结构化出题，文本 + 视觉双模型审校与确定性质量门禁；学生端七种题型互动、分层提示与错题多轮陪练闭环；FastAPI + PostgreSQL JSONB + React 19
用于系统综述与 meta 分析的 R 脚本，可从 DOI 提取全面的元数据。R scripts for extracting comprehensive metadata from DOIs for systematic reviews and meta-analyses
面向私有仓库与本地模型的自托管、厂商中立编码 Agent。TUI 优先，支持 16 家提供商的 BYOK 或通过 Ollama/LM Studio 完全离线运行。每个 Agent 操作均经过沙箱隔离、日志记录与签名——构建为可验证，而非仅可信任。Self-hosted, vendor-neutral coding agent for private repos and local models. TUI-first, BYOK across 16 providers or fully offline with Ollama/LM Studio. Every agent action is sandboxed, logged, and signed — built to be verified, not just trusted.
Warzone2100 的 Skirmish AI 机器人Skirmish AI bot for Warzone2100
AutoScientists：确定性的多 agent 科学发现协调框架 —— docxology/template 可复现研究范式的公开范例（测试覆盖率 ≥90%，无 mock，GitHub 与 Zenodo 10.5281/zenodo.20533669 双重发布）。AutoScientists: a deterministic multi-agent scientific-discovery coordination harness — public exemplar of the docxology/template reproducible-research paradigm (≥90% test coverage, no mocks, double-published GitHub + Zenodo 10.5281/zenodo.20533669).
比特币元宇宙——主权数字土地上的主权 AI Agent。The Bitcoin Metaverse — Sovereign AI agents on sovereign digital land.
Mistral Studio 产品营销策略、品类叙事、定位与信息框架、从专业消费者到企业的分层、销售作战卡、品牌资产与市场分析。Mistral Studio Product Marketing Strategy, Category Narrative, Positioning & Messaging Framework, Prosumer-to-Enterprise Segmentation, Sales Battlecards, Brand Assets & Market Analysis.
MetaGráfica 是一种面向出版级二维矢量图形的描述语言——从纯文本编译为 EPS、SVG 或 PDF，适配科学图表与绘图，由单一自包含编译器完成。MetaGráfica is a descriptive language for publication-quality 2D vector graphics — scientific figures and plots compiled from plain text to EPS, SVG or PDF by a single self-contained compiler.
围绕病理 foundation models（尤其是 Prov-GigaPath）的文献综述与可复现性研究Literature review and reproducibility work on pathology foundation models, with a focus on Prov-GigaPath.
用于科学论文写作的 AI agents，可在评论中给出修正意见，使科学写作准确、流畅、无误，适合在科学期刊上发表。AI agents for Scientific Manuscript Writing comments, giving out correction comments to make scientific writing accurate polished without errors and suitable for publications in scientific journals.
一道 Python 守门机制：在不分类提示词的前提下，阻止不可信的 agent 数据去构造敏感的工具调用参数。A Python gate that prevents untrusted agent data from authoring sensitive tool-call arguments—without classifying prompts.
由文本挖掘驱动的科学文献综述。A text mining-driven review of scientific literature.
夏季开放与可复现研究（SOAR²）训练营的项目官网。Program Website for the Summer Open And Reproducible Research (SOAR²) Camp
支撑 AV/CAV 交通拥堵缓解半系统综述的数据与代码。Data and coding supporting a semi-systematic review of AV/CAV congestion mitigation.
AI-Native 钢琴学习 RAG 助手——在线且免费（GitHub Pages -> Google Cloud Run -> Neon pgvector -> Groq）：具备引用、护栏与内置可观测性（延迟/token/成本）的摄取与查询流水线（rewrite -> 混合检索/RRF -> rerank -> LLM），数据持久化至可检索数据库。Node.js 实现，不依赖付费 API。AI-Native Piano Learning RAG Assistant - live & free (GitHub Pages -> Google Cloud Run -> Neon pgvector -> Groq): ingestion + query pipeline (rewrite -> hybrid search/RRF -> rerank -> LLM) with citations, guardrails, and built-in observability (latency/tokens/cost) persisted to a searchable DB. Node.js, no paid APIs.
生产级 LLM 网关：兼容 OpenAI 的 API，支持语义缓存，可在 Claude/OpenAI/Ollama 之间进行智能路由与成本分析。Production LLM gateway: OpenAI-compatible API, semantic caching, intelligent routing across Claude/OpenAI/Ollama, cost analytics.
HateMirage——可解释的伪仇恨检测与多维推理，ICON 2026 共享任务。在 HateMirage 语料库（4,530 条带标注的伪仇恨评论）上进行目标识别 + 意图与隐含意义生成。HateMirage - Explainable Faux Hate Detection and Multi-Dimensional Reasoning - Shared Task @ ICON 2026. Target identification + Intent and Implication generation over the HateMirage corpus (4,530 annotated Faux Hate comments).
端到端 Python RAG 框架，包含文档摄取、语义搜索、对话式 AI、多用户检索、可观测性、结构化输出与评估。An end-to-end Python RAG framework featuring document ingestion, semantic search, conversational AI, multi-user retrieval, observability, structured outputs, and evaluation.
Windows 上 Claude Code 的计算机使用能力：一个 MCP server，让 Claude 能够看到你的屏幕并操控鼠标和键盘，对齐 Anthropic 的 computer-use 动作集。Computer use for Claude Code on Windows: an MCP server that lets Claude see your screen and drive mouse and keyboard, mirroring Anthropic's computer-use action set
Agentic-SDLC 审计工具，整个工具完全通过其所记录的流水线自身构建：规划、agent 构建、对抗式 AI 审查、CI 卡点、人工授权合并。针对 AI agent 开发规范输出确定性评分报告。Agentic-SDLC audit tool built entirely through the pipeline it documents: plan, agent build, adversarial AI review, CI gate, human-authorized merge. Deterministic scored reports on AI-agent development discipline.
无凭据 MCP 策略代理，具备细粒度能力与防篡改凭证。Secretless MCP policy proxy with scoped capabilities and tamper-evident receipts
接收研究问题 → 检索网页 → 阅读关键页面 → 综合输出带引文与置信度分数的结构化报告。Takes a research question → searches the web → reads key pages → synthesizes a structured report with citations and confidence scores
面向编程 Agent 的确定性业务变更安全闸门——基于证据的 BLOCK/ALLOW 决策。Deterministic business-change safety gate for coding agents — evidence-backed BLOCK/ALLOW decisions
集成 ChatGPT web 与 Codex，用于 Agentic 编码工作流：ChatGPT 负责思考，Codex 负责执行，以节省 Codex 用量配额。Integrates ChatGPT web and Codex for agentic coding workflows ChatGPT thinks, Codex executes for saving Codex usage limits.
来自 Principia 的研究问题合集Collective gatherings of research questions from Principia
🧠 为 Telegram、Discord 与 Reddit 打造的私有、源码溯源知识库，基于混合 RAG，并提供每周摘要。🧠 Private, source-grounded knowledge base for Telegram, Discord, and Reddit with hybrid RAG and weekly digests.
面向 Codex 的可溯源科学 PDF 阅读 skill，支持来源绑定的注释、细致的图表与公式讲解以及上下文证据问答Traceable scientific PDF reader skill for Codex with source-bound annotations, detailed figure and formula teaching, and contextual evidence questions.
插入到 MCP 服务器前方的安全与可信代理——策略拦截、rug-pull 检测、JIT 审批、哈希链审计。124 个测试。MCP 서버 앞에 끼어드는 보안·신뢰성 프록시 — 정책 차단, rug-pull 탐지, JIT 승인, 해시 체인 감사. 124 tests
综述基于性状的方法以量化农业对土壤无脊椎动物影响的研究A systematic review of trait-based approaches to quantify agricultural impacts on soil invertebrates
Microwave Method：agent 工厂 + 受治理的 LLM wiki。每个 agent 一次性重型处理，每个 feature 轻量处理。上下文只"烹饪"一次，按缓存价"回热"。Microwave Method: agent factory + governed LLM wiki. Heavy pass once per agent, light pass per feature. Context is cooked once, reheated at cache price.
开源 Python 平台，用于粗糙随机波动率的模拟、定价与统计分析。索尔福德大学独立研究项目。Open-source Python platform for simulation, pricing, and statistical analysis of rough stochastic volatility. Independent research programme, University of Salford
端到端系统化交易平台，覆盖可复现研究、共享行情与成交合约、投资组合回测与实盘交易。End-to-end systematic trading platform spanning reproducible research, shared market data and execution contracts, portfolio backtesting, and live trading.
基于 Zotero 文献库的每日 arXiv 论文推荐。Daily arXiv paper recommendations based on Zotero library
一个轻量级、本地优先的桌面文献管理与引用工具，支持学术文献、元数据、笔记以及 BibTeX/BibLaTeX 工作流A lightweight, local-first desktop bibliography and citation manager for scholarly references, metadata, notes, and BibTeX/BibLaTeX workflows.
基于 Transformer 架构的可复现研究原型，用于长周期建筑能耗预测，结合可微的热物理约束并支持跨建筑泛化A reproducible research prototype for long-horizon building energy forecasting using transformer architectures with differentiable thermal-physics constraints and cross-building generalization.
von der Heyde, L., Keusch, F., Buskirk, T. D., & Eck, A. (2026). AI in the Loop?! A Systematic Review of the Use of Large Language Models in Survey and Public Opinion Research. SocArXiv. https://doi.org/10.31235/osf.io/eubj4_v1 的复现材料与文献数据库Replication materials and literature databases for von der Heyde, L., Keusch, F., Buskirk, T. D., & Eck, A. (2026). AI in the Loop?! A Systematic Review of the Use of Large Language Models in Survey and Public Opinion Research. SocArXiv. https://doi.org/10.31235/osf.io/eubj4_v1.
对接日本 gBizINFO 企业信息 REST API v2 的 MCP serverMCP server for Japan’s gBizINFO corporate information REST API v2.
用于搜索与检索日本官方 e-Stat 政府统计数据的 MCP serverMCP server for searching and retrieving Japan’s official e-Stat government statistics.
Claude Code、OpenCode、Cursor 与 Vercel skills CLI 可复用的 Agent skill。Reusable agent skills for Claude Code, OpenCode, Cursor, and the Vercel skills CLI.
SAM 的纯文档概述。SAM 是一个由 9 个 agent 组成的 LLM pipeline，用于自动化系统综述与 meta 分析。实现位于私有仓库。Documentation-only overview of SAM, a 9-agent LLM pipeline that automates systematic reviews and meta-analyses. Implementation lives in a private repository.
基于 Python GUI 的应用，用于自动化稳定同位素研究中质谱数据的归一化、处理与统计分析A Python GUI application built to automate the normalization, processing, and statistical analysis of mass spectrometer data for stable isotope research.
面向科学与医学写作的人性化润色的阿英双语 skill，同时保持学术准确性。Bilingual Arabic-English skill for humanizing scientific and medical writing while preserving scholarly accuracy.
为仿制药临床法规事务团队打造的法规准备加速器。A regulatory prep accelerator for a generic-drug Clinical Regulatory Affairs team.
集成 RAG、知识图谱、语义搜索、论文推荐与本地 LLMs 的 AI 研究助手AI Research Assistant with RAG, knowledge graphs, semantic search, paper recommendations, and local LLMs.
确定性研究亲和度匹配、文献空白发现（Ω 指数）、跨学科共同作者雷达，以及不产生生成式幻觉的逐字引文证据审计。Deterministic research affinity matching, literature white-space discovery (Ω index), cross-disciplinary co-author radar, and verbatim citation evidence auditing without generative hallucinations.
Codex skill，面向交通运输与低空出行领域的学术写作（知识蒸馏自 LT）。Codex skill for transportation and low-altitude mobility academic writing (Knowledge distillation from LT)
一个可复用的 skill，用于对临床研究、系统综述和 meta-analyses 进行严谨且有来源支持的评估与分析。A reusable skill for rigorous, source-backed appraisal and analysis of clinical studies, systematic reviews, and meta-analyses.
EconOps：面向经济学实证研究的透明、可复现研究工作流EconOps is a reproducible research workflow for transparent empirical work in economics.
模板化的元研究范例：自文档化的可复现研究仓库。Template meta-research exemplar: self-documenting reproducible research repo
The Triplicate：数据驱动的大幅面报纸排版引擎 —— docxology/template 可复现研究范式的公开范例（测试覆盖率 ≥90%，无 mock，GitHub 与 Zenodo 10.5281/zenodo.20533675 双重发布）。The Triplicate: a data-driven large-format newspaper layout engine — public exemplar of the docxology/template reproducible-research paradigm (≥90% test coverage, no mocks, double-published GitHub + Zenodo 10.5281/zenodo.20533675).
使用 Kaggle 数据集分析屏幕时间对生产力多方面影响的研究问题This uses a dataset from Kaggle, to analyze multiple research questions on how screen time affects productivity in multiple aspects.
我的 GitHub 个人主页，涵盖生物信息学、结构建模与可复现性研究Bioinformatics, structural modelling, and reproducible research — my GitHub profile.
中文學術寫作的後設論述量尺 · A descriptive metadiscourse scale for Chinese academic writing (not a detector)
你的 Mac 又慢又烫又吵。machogs 用朴实的英文告诉你罪魁祸首是谁，揪出那些吞噬 CPU 的 AI 工具辅助进程和卡住的进程。Your Mac is slow, hot and loud. machogs tells you what is doing it, in plain English. Finds the AI tool helpers and stuck processes eating your CPU.
可衡量的 Agent 分诊：判断一次编辑是否需要人工，还是可由 AI 处理。我们衡量 Agent 下方的图能否触达守住变更的测试（7 个仓库合并值 0.42），并发布 fail-closed 闸门，将未经证明的映射路由到人工核验。CLI · MCP · SARIF · 内置于 HydraDB。Agent triage, measured: does this edit need a human, or can the AI handle it? We measure whether the graph under the agent can reach the tests that guard a change (0.42 pooled, 7 repos) and ship the fail-closed gate that routes unproven maps to human verification. CLI · MCP · SARIF · in-engine on HydraDB.
一个研究级 Python 流水线，用于分析 Wireshark 抓包数据，提取网络流特征，进行统计分析、异常检测，并生成可用于 GAN 入侵检测系统的机器学习就绪数据集A research-grade Python pipeline for analyzing Wireshark packet captures, extracting network flow features, performing statistical analysis, anomaly detection, and generating machine learning-ready datasets for GAN-based Intrusion Detection Systems.
面向 AI 编程助手的 Agent skills —— 文档解析、PDF 工程、系统综述研究、闪卡与 WhatsApp 自动化。兼容 Claude Code、Cursor、Codex、ZCode。Agent skills for AI coding assistants — document parsing, PDF engineering, systematic-review research, flashcards, and WhatsApp automation. Works with Claude Code, Cursor, Codex, ZCode.
基于 Docker Compose 的 AI、LLM 与 RAG 平台基础设施蓝图，含路由、存储与可观测性说明。AI, LLM, and RAG platform infrastructure blueprint with Docker Compose, routing, storage, and observability notes.