duyou.me | 独有·我

坚持每一天，做更好的自己

企业的 AI 技能递送者

“AI 懂你的业务，却无法触碰你的系统？我们负责连接。”

eBay 技术负责人 · 20年金融系统和AI赋能经验

持有两项美国机器学习专利

分布式系统架构 · 实时流处理 · 支付风控
图神经网络 · LLM

SOLO项目 AI 前沿情报开源热点

项目里程碑

公众号上线

2026年1月

「谢先生的AI深析札记」创刊，专注 AI 技术深度解析与行业洞察

「谢先生的AI读典实录」视频号开播，AI 驱动的经典解读

视频号运营

2026年2月

硬件升级

2026年3月

购置 MacBook Pro M5 Max 64GB，搭建本地 AI 研究工作台

督友小程序上线

2026年4月

强迫症式轻社群自律监督小程序，坚持每一天，遇见更好的自己

duyou.me 首页发布，展示项目与研究成果

域名上线

2026年4月

博客系统上线

2026年4月

Solo 日志博客系统发布，记录一人公司的成长与思考

公众号文章向系列化与深度化转型，首重技术深度；「谢先生的AI读典实录」《黄帝内经》系列播放量突破 1,000。

媒体运营升级

2026年5月

业务与生产力升级

2026年5月

启动 AI 跨境电商项目探索；正式订阅启用 Kimi 专业版，提升团队生产力。

AI 情报页面升级

2026年7月

重构 briefings.astro 情报详情页，优化 Scout Agent 分类展示与外部来源链接，全面提升 AI 前沿情报的阅读体验

《拉普拉斯的盲点》第一卷《地球囚笼》在番茄小说完本，点击阅读

番茄小说完本

2026年8月

iOS 应用上线

2026年8月

万事宝与如意宝 iOS 应用正式上线，万事宝（生活实用） / 如意宝（精神疗愈）

「谢先生的AI读典实录」与「谢先生的AI深析札记」进入从完成到完善、再到完美的持续打磨阶段；完成《拉普拉斯的盲点》后续各卷；深化 iOS 应用的体验深度与趣味性；探索 AI 视频长短剧新赛道

持续创作

进行中

核心成果

最后更新：2026年8月18日

AI 研究成果

公众号「谢先生的AI深析札记」已发布文章，涵盖产业分析、Agent 架构、本地部署实战。

→ 中国 AI 产业格局一超多强、四龙夺珠

→ Agent 思考模式智能体决策链设计

进入研究库 →

督友小程序

强迫症式轻社群自律监督应用。采用混合云架构，支持离线网络缓存自动降级，打造无缝体验。

→ 架构设计API 优先 + 本地回退

→ 离线支撑断网环境无缝打卡

查看详情 →

企业 AI 技能递送

与其支付高昂的人力完成繁琐录入，不如交付一套“听话、高效、不出错”的自动化流程。让 AI 真正接管您的老旧系统与专业软件。

● 降本增效比雇佣初级文员更划算

● 落地闭环交付不休息的自动化生产力

● 安全合规数据不出内网，账号原生安全

查看详情 & 深度演示 →

Agent Gateway 架构

从简单 Proxy 到工业级 Agent Gateway：通过 LiteLLM 构建分层容灾与多级重试架构，让 Claude Code 完美运行在 NVIDIA NIM 免费额度上。

● 分层容灾 Tier 1 高智力 + Tier 2 高稳定兜底

● 工具清洗 Schema Enforcement 过滤幻觉参数

● 永不宕机层内重试 + 层间切换保障稳定性

阅读完整文章 →

Harness Engineering 系列

2026 年 AI 工程师的新必修课。从入门到精通，再到生产级实战，全面解析如何给 Agent 装上缰绳，让模型能力变得可靠、可控、可评估。

● 从入门到精通控制循环、工具路由、护栏、记忆系统

● 从精通到大师长周期任务、容错自愈、成本控制

● 生产级架构五层稳定架构、神经符号分离

● 最新：记忆工程 Agent 失忆问题与解决方案（5月12日）

阅读完整系列 →

小说

科幻长篇小说《拉普拉斯的盲点》第一卷《地球囚笼》在番茄小说完本，构建基于拉普拉斯妖思想的硬科幻世界观。

→ 第一卷完本《地球囚笼》

→ 连载平台番茄小说

阅读小说 →

苹果 App

万事宝与如意宝 iOS 应用正式上线，覆盖生活实用与精神疗愈两大场景。

→ 万事宝生活实用助手，万事如意

→ 如意宝精神疗愈陪伴，事事如意

万事宝 → 如意宝 →

支付与财务 AI 深度研究

深耕支付行业 20 年。专注于支付风控、金融科技合规以及企业财务 AI 化转型的深度融合探索。

● 研究预研中 · 2026 Q3 发布

敬请期待

Core Achievements

Last updated: August 18, 2026

AI Research

Articles published on "Mr. Xie's AI Deep Analysis Notes," spanning industry analysis, Agent architecture, and on-premises deployment.

→ China AI Landscape One giant, many contenders

→ Agent Reasoning Decision-chain design for intelligent agents

Explore Research →

Duyou Mini Program

An OCD-friendly lightweight community for self-discipline accountability. Hybrid cloud architecture with offline cache auto-fallback for a seamless experience.

→ Architecture API-first + local fallback

→ Offline Resilience Seamless check-in without network

Learn More →

Delivering AI Skills to Enterprise

Rather than paying steep labor costs for tedious data entry, deliver an automated process that is obedient, efficient, and error-free. Let AI truly take over your legacy systems and professional software.

● Cut Costs More economical than hiring junior staff

● End-to-End Delivery Non-stop automated productivity

● Secure & Compliant Data stays on-premises, native account safety

Learn More & Deep Demo →

Agent Gateway Architecture

From Simple Proxy to Industrial-grade Agent Gateway: Build tiered resilience and multi-level retry architecture with LiteLLM, running Claude Code flawlessly on NVIDIA NIM free credits.

● Tiered Resilience Tier 1 high-intelligence + Tier 2 stability fallback

● Tool Sanitization Schema Enforcement filters hallucinated parameters

● Never Down Intra-tier retry + inter-tier switching ensures stability

Read Full Article →

Harness Engineering Series

The new essential skill for AI engineers in 2026. From beginner to master, then to production实战, comprehensively explaining how to put reins on Agents, making model capabilities reliable, controllable, and evaluable.

● Beginner to Master Control loops, tool routing, guardrails, memory systems

● Master to Grandmaster Long-running tasks, fault tolerance, cost control

● Production Architecture Five-layer stable architecture, neuro-symbolic separation

● Latest: Memory Engineering Agent amnesia problem & solutions (May 12)

Read Full Series →

Novel

Completed Book One: Earth Cage of the sci-fi novel "Laplace's Blind Spot" on Fanqie Novel, building a hard sci-fi world around Laplace's demon.

→ Book One Complete Earth Cage

→ Platform Fanqie Novel

Read the Novel →

Apple Apps

Launched WanShiBao and Ruyibao iOS apps, covering practical life tools and spiritual wellness.

→ WanShiBaoPractical life assistant

→ RuyibaoSpiritual wellness companion

WanShiBao → Ruyibao →

Payment & Finance AI Research

20 years immersed in the payment industry. Focused on payment risk control, fintech compliance, and the deep convergence of enterprise finance and AI transformation.

● Research in progress · Q3 2026 release

Coming Soon

AI 前沿周报

最后更新：2026年09月03日

Introducing Brand Studio: The creative production platform powered by your brand

2026-09-03

We’re excited to introduce Brand Studio by Stability AI, the end-to-end creative production platform powered by your brand.

Stability AI News

Sharing a new way to work with Stable Audio

2026-09-03

We’re sharing two new ways for you to use Stable Audio 3.0. We now have a DAW plugin and a more advanced generation experience onStableAudio.com.

Stability AI News

An embodied, privacy-first AI agent that can control mobile apps via GUI

2026-09-03

Read the full article: An embodied, privacy-first AI agent that can contr...

Hacker News AI

DiDrive: A Risk-Aware Hierarchical Diffusion Framework for Safe Offline Reinforcement Learning in Autonomous Driving

2026-09-03

arXiv:2609.01609v1 Announce Type: new Abstract: While diffusion models effectively capture multimodal behavioral priors for autonomous driving, offline reinforcement learning (RL) policies remain susc...

Arxiv (AI/LG)

Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives

2026-09-03

arXiv:2609.01736v1 Announce Type: cross Abstract: Large language models (LLMs) augmented with external tools have demonstrated remarkable capability in solving complex real-world tasks. However, exist...

Arxiv (CS/MA)

SpeakPay: Domain-Adaptive LoRA Fine-Tuning of Whisper for Low-Resource Nepali Financial Speech Recognition

2026-09-03

arXiv:2609.01737v1 Announce Type: new Abstract: Mobile payment applications in Nepal are graphically mediated and largely inaccessible to visually impaired users. This paper presents SpeakPay, a voice...

Arxiv (AI/CL)

查看全部周报

开源热点周报

由 GitHub Scout Agent 自动追踪的最具潜力的开源项目

最后更新：2026 年 9 月 3 日

本周 GitHub 热点继续围绕一个核心主题：

1. 把大语言模型压缩成可本地运行、可审计的个人工具。从 agentic IDE 到多模型编排层，开源社区正在把"智能体"从 demo 变成基础设施。对一人公司而言，这意味着可以把昂贵的工程团队压缩成一套本地优先、零信任的生产栈。 2. AI 输出的"最后一公里"体验成为新战场。去 AI 味改写、材质插画、3D 代码库可视化等工具说明：能生成已不够，关键是输出是否自然、可用、能否直接嵌入工作流。

不要追逐最新模型版本或越狱提示词。真正的杠杆在于：选一个熟悉的工具链，把重复决策自动化，把模型能力封装成可复用的个人工作流。

⭐ XiaoDuoYa/codex-with-chatgpt：ChatGPT thinks

⭐ 2296

AI 智能体与基础设施

ChatGPT thinks. Codex works. Use ChatGPT as the planning brain while keeping the Codex harness.

GitHub →

⭐ cbrock84/headcount：An 智能体 organization for Claude Code, structured as a company — 15+

⭐ 1126

开发者工具

An 智能体 organization for Claude Code, structured as a company — 15+ departments, 125+ skills, each independently installable.

GitHub →

⭐ jlrouzies-fr/DLSS5-Feeder：DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without

⭐ 611

开发者工具

DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without any DLSS — feeds a synthetic DLAA contract (ReShade depth + motion vectors) to the DLSS 5 add-on via a private D3D12 device.

GitHub →

⭐ 2akouwu/reverify：Verified reverse engineering: AI RE grounded on deterministic tools -

⭐ 599

开发者工具

Verified reverse engineering: AI RE grounded on deterministic tools - results checked against the binary, not hallucinated.

GitHub →

⭐ Nanako0129/sepia：De-AI writing skill for any 智能体 Skills-compatible 智能体 (77+ via the

⭐ 1666

AI 内容人化与可视化

De-AI writing skill for any 智能体 Skills-compatible 智能体 (77+ via the Skills CLI), with native plugins for Claude Code, Codex, Grok Build, and Antigravity. Narrative-architecture repair for fiction, venue-matched rules for professional prose. Based on StoryScope (arXiv:2604.03136).

GitHub →

⭐ GangTailorUpgrade/undress-service：Dress AI Sponsor

⭐ 945

AI 内容人化与可视化

Dress AI Sponsor

GitHub →

查看全部开源速递

Open Source Weekly

Most promising open-source projects, auto-tracked by GitHub Scout Agent

Last updated: September 03, 2026

本周 GitHub 热点继续围绕一个核心主题：

1. 把LLM压缩成可本地运行、auditable的个人工具。从 agentic IDE 到multi-modelorchestration层，open source社区正在把"agent"从 demo 变成infrastructure。对一人公司而言，这意味着可以把昂贵的工程团队压缩成一套local-first、zero-trust的生产栈。 2. AI 输出的"最后一公里"体验成为新战场。de-AI-ingrewriting、material illustration、3D codebasevisualization等工具说明：能generation已不够，关键是输出是否自然、可用、能否直接embeddingworkflow。

不要追逐最新model版本或越狱prompt。真正的杠杆在于：选一个熟悉的工具链，把重复决策automation，把model能力封装成可复用的个人workflow。

⭐ XiaoDuoYa/codex-with-chatgpt: ChatGPT thinks

⭐ 2296

AI Agents & Infrastructure

ChatGPT thinks. Codex works. Use ChatGPT as the planning brain while keeping the Codex harness.

GitHub →

⭐ cbrock84/headcount: An agent organization for Claude Code, structured as a company — 15+ departments, 125+ skills, each independently

⭐ 1126

Developer Tools

An agent organization for Claude Code, structured as a company — 15+ departments, 125+ skills, each independently installable.

GitHub →

⭐ jlrouzies-fr/DLSS5-Feeder: DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without any DLSS — feeds a synthetic DLAA contract (ReShade

⭐ 611

Developer Tools

DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without any DLSS — feeds a synthetic DLAA contract (ReShade depth + motion vectors) to the DLSS 5 add-on via a private D3D12 device.

GitHub →

⭐ 2akouwu/reverify: Verified reverse engineering: AI RE grounded on deterministic tools - results checked against the binary, not

⭐ 599

Developer Tools

Verified reverse engineering: AI RE grounded on deterministic tools - results checked against the binary, not hallucinated.

GitHub →

⭐ Nanako0129/sepia: De-AI writing skill for any Agent Skills-compatible agent (77+ via the Skills CLI), with native plugins for Claude Code,

⭐ 1666

AI Content Humanization & Visualization

De-AI writing skill for any Agent Skills-compatible agent (77+ via the Skills CLI), with native plugins for Claude Code, Codex, Grok Build, and Antigravity. Narrative-architecture repair for fiction, venue-matched rules for professional prose. Based on StoryScope (arXiv:2604.03136).

GitHub →

⭐ GangTailorUpgrade/undress-service: Dress AI Sponsor

⭐ 945

AI Content Humanization & Visualization

Dress AI Sponsor

GitHub →

Explore All Projects

duyou.me | 独有·我

duyou.me | Duyou·Me

项目里程碑

公众号上线

视频号运营

硬件升级

督友小程序上线

域名上线

博客系统上线

媒体运营升级

业务与生产力升级

AI 情报页面升级

番茄小说完本

iOS 应用上线

持续创作

Project Milestones

WeChat Official Account

Video Channel

Hardware Upgrade

Duyou Mini Program Launch

Site Launched

Blog System Launch

Media Operations Upgrade

Business & Productivity

AI Intelligence Page Overhaul

Fanqie Novel Published

iOS Apps Launched

Keep Building

核心成果

AI 研究成果

督友小程序

企业 AI 技能递送

Agent Gateway 架构

Harness Engineering 系列

小说

苹果 App

支付与财务 AI 深度研究

Core Achievements

AI Research

Duyou Mini Program

Delivering AI Skills to Enterprise

Agent Gateway Architecture

Harness Engineering Series

Novel

Apple Apps

Payment & Finance AI Research

AI 前沿周报

Introducing Brand Studio: The creative production platform powered by your brand

Sharing a new way to work with Stable Audio

An embodied, privacy-first AI agent that can control mobile apps via GUI

DiDrive: A Risk-Aware Hierarchical Diffusion Framework for Safe Offline Reinforcement Learning in Autonomous Driving

Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives

SpeakPay: Domain-Adaptive LoRA Fine-Tuning of Whisper for Low-Resource Nepali Financial Speech Recognition

AI Frontiers Weekly

Introducing Brand Studio: The creative production platform powered by your brand

Sharing a new way to work with Stable Audio

An embodied, privacy-first AI agent that can control mobile apps via GUI

DiDrive: A Risk-Aware Hierarchical Diffusion Framework for Safe Offline Reinforcement Learning in Autonomous Driving

Harness Engineering in LLM Tool Use via Agent-Native Reusable Tool Primitives

SpeakPay: Domain-Adaptive LoRA Fine-Tuning of Whisper for Low-Resource Nepali Financial Speech Recognition

开源热点周报

⭐ XiaoDuoYa/codex-with-chatgpt：ChatGPT thinks

⭐ cbrock84/headcount：An 智能体 organization for Claude Code, structured as a company — 15+

⭐ jlrouzies-fr/DLSS5-Feeder：DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without

⭐ 2akouwu/reverify：Verified reverse engineering: AI RE grounded on deterministic tools -

⭐ Nanako0129/sepia：De-AI writing skill for any 智能体 Skills-compatible 智能体 (77+ via the

⭐ GangTailorUpgrade/undress-service：Dress AI Sponsor

Open Source Weekly

⭐ XiaoDuoYa/codex-with-chatgpt: ChatGPT thinks

⭐ cbrock84/headcount: An agent organization for Claude Code, structured as a company — 15+ departments, 125+ skills, each independently

⭐ jlrouzies-fr/DLSS5-Feeder: DLSS 5 neural rendering in D3D11/D12/Vulkan games that ship without any DLSS — feeds a synthetic DLAA contract (ReShade

⭐ 2akouwu/reverify: Verified reverse engineering: AI RE grounded on deterministic tools - results checked against the binary, not

⭐ Nanako0129/sepia: De-AI writing skill for any Agent Skills-compatible agent (77+ via the Skills CLI), with native plugins for Claude Code,

⭐ GangTailorUpgrade/undress-service: Dress AI Sponsor