周报 · 2026 年第 41 周
GitHub 新星 · 汇总该周新收录项目
当周新增收录 318 个项目
覆盖日期：2026-10-05 至 2026-10-11 · 北京时间
发布时间：2026/10/14 14:30:30
本刊有 2 个项目已不可公开，相关条目已隐藏。
部分来源已不可公开，本期文章将按其余项目修订后重新展示。下方为已保存的项目资料。
本期项目资料
收录数按本期首次达标收录的项目统计。 本期最早收录记录为 2026/10/9 11:44:24，更早时段没有保存的收录记录。
- storytold/photocraftRust · 58,739 Stars
图像编辑、PSD编辑
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
PhotoCraft: An open-source, photo editor written by scratch in pure Rust
编报时的 AI 中文解读
PhotoCraft 是一款用纯 Rust 编写的开源照片编辑器，旨在以原生应用形式重新实现 Adobe Photoshop 的核心功能，支持图层、蒙版、调整图层、图层样式、文字、矢量、画笔以及真实的 PSD 文件。它面向熟悉 Photoshop 的用户，提供桌面和网页版本，并支持离线使用。
编报时的 README 依据
Open, edit and save layered Photoshop documents. Re-saving keeps the render of 307 of the 309 psd-tools test files. </td> <td width="25%" valign="top"> <h3>🤖 Agent-ready</h3> Every action is a command, so you can drive the same engine from the UI, the CLI, a JSON control channel or an MCP server. </td> </tr> </table> <br> ## Features Every screenshot here is the real app at work on public-domain art, rendered offscreen through its control channel. <table> <tr> <td width="50%" valign="top">
Every menu item, tool and dialog runs a command from one registry of 500+ commands. The UI, the CLI, the JSON control channel and the MCP server all call the same commands, so anything you can click, a script or an AI agent can do too. ```sh # Headless: open, edit, save photocraft-cli run wave.psd \ --cmd filter.sharpen.smartSharpen --params '{"amount":80}' \ --cmd layer.newAdjustmentLayer.curves --params '{"points":[[0,0],[64,48],[192,212],[255,255]]}' \ --out wave-final.png # Apply one action list to a folder of images
- **Clean-room:** implemented from public specs and observed behaviour only. No proprietary code, shaders or assets. - **Tested:** more than 1,700 tests, including PSD round trips, synthetic generators, compositor oracles and multi-depth checks. ## Get started ```sh git clone https://github.com/storytold/photocraft cd photocraft cargo run --release -p photocraft -- image.psd # the desktop app cargo test --workspace # the test suite ```
Installers for macOS, Windows, Linux, FreeBSD and the web are attached to each [GitHub release](https://github.com/storytold/photocraft/releases). On Linux you can pick an AppImage, a `.deb`, an `.rpm`, a tarball or a Flatpak bundle. The bundle needs the freedesktop runtime from [Flathub](https://flathub.org/setup), which `flatpak` offers to install along with it: ```sh flatpak install --user photocraft-<version>-linux-x86_64.flatpak # or -linux-aarch64 flatpak run ai.storyteller.photocraft ```
> **Status:** PhotoCraft is in early alpha, and we want to be straight about where it stands: much of Photoshop's feature surface exists in some form, but **it is not yet a Photoshop replacement for daily professional work**. The biggest gaps are AI/generative features, about twenty missing tools, depth in typography and pro workflows, and plug-in compatibility. Every Photoshop menu item is wired to a command ([`docs/parity.md`](docs/parity.md)), but that measures wiring, not behaviour. The honest, dimension-by-dimension picture and where we're going next are in the [roadmap's parity
- NandhaKishorM/layaPython · 32,807 Stars
文本分类、文本评分、模型本地部署
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
编报时的 AI 中文解读
Laya 是一个多语言、非自回归的决策引擎，用于对文本进行类型化决策（如选择、评分、是非判断），支持100多种语言，通过路由器选择适当的检查点，适用于客服工单分类、意图识别等场景。
编报时的 README 依据
Installing the package also installs a `laya` command for quick local testing, no script needed: ```bash laya "I was charged twice, please refund" # routing decision only; works offline, no download laya "Refactor this service" --predict # full answers (downloads the checkpoint on first use) laya "Mein Konto wurde zweimal belastet" --lang de # force a language instead of detecting it laya "My payment failed twice" --lang-guess en # a soft hint: nudge routing, still fall through to detection
laya "My payment failed twice" --model ml # pin a checkpoint: names, aliases and casing all resolve as the SDK resolves them laya "My payment failed twice" --preset triage # answer a ready-made preset (triage, email, guard, moderation, router) laya --batch tickets.txt --predict # score a file of requests, one per line, in one batch cat tickets.txt | laya --batch - --predict --json # stdin; one JSON line of answers per request laya "Where is my card" --questions intents.json # answer your own questions, written in a JSON file
Python 3.10 or newer. Optional extras: `laya[serve]` (HTTP server), `laya[mcp]` (MCP server), `laya[langchain]` (LangChain and LangGraph), `laya[llamaindex]` (LlamaIndex selectors), `laya[crewai]` (CrewAI routing), `laya[onnx]` (ONNX Runtime), `laya[fast]` (TileLang GPU fast path), `laya[structured]` (pydantic models in `decide`). Step-by-step setup for each platform, CPU-only or GPU PyTorch builds, and troubleshooting are in [Installation details](#installation-details).
If your application is a uv project, add Laya as a dependency instead: ```bash uv add laya uv run python -I -c "import laya; print(laya.__version__)" ``` `--torch-backend` applies to `uv pip` only; in a uv project, uv's PyTorch guide shows how to set the PyTorch index in `pyproject.toml`. **Model setup and troubleshooting** Continue with the [Router quickstart](#quickstart-route-mode-recommended) to run inference. Loading a Hub checkpoint requires access to Hugging Face on its first download; the quickstart's `Router(preload=True)` loads all three configured checkpoints at construction.
> In the table above, 16 to 18 of 20 requests were answered correctly with up to about 4,000 tokens of text before them; beyond that results vary (8 to 17 of 20), so check long-document accuracy on your own data. Short inputs give identical answers with `max_len=8192`, and speed follows the input's real length, not the limit: short inputs are unchanged, and a 4,000-token input takes about 1.7 s on an Apple GPU. Name the checkpoint with `model="multilingual"`, since long mostly-English text would otherwise route to the English checkpoint. ```python from laya import Router
- Niko1221/StrataC++ · 23,946 Stars
模型本地部署、模型接口服务、聊天辅助
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.
编报时的 AI 中文解读
Strata 是一个在普通消费级 PC 上运行大型 AI 模型的工具，支持聊天、写代码、读图，并提供本地 API 服务。适用于拥有 12GB 以上显存显卡的个人用户，无需服务器即可本地运行。
编报时的 README 依据
<p align="center"><img src="docs/media/runpagoda.png" width="900" alt="The Strata app's Monitor tab next to a coding agent"><br> <sub>The Strata app's <b>Monitor</b> (left) while a coding agent writes the pagoda garden from the video (right)</sub></p> - **In the browser:** open `http://127.0.0.1:8080`. It has **Chat**, a live **Monitor** of the model and your GPU/CPU/RAM, and **About** with the settings and addresses. - **Your apps and coding agents:** add an "OpenAI-compatible" provider with the base URL **`http://127.0.0.1:8080/v1`**. Any API key and any model name work.
- Apps that use Anthropic's API: `http://127.0.0.1:8080/v1/messages` (Claude Code: `ANTHROPIC_BASE_URL=http://127.0.0.1:8080`). - Codex CLI and other apps that use the OpenAI Responses API: `/v1/responses` ([setup](docs/DETAILS.md#the-responses-api-and-codex-cli)). - **Thinking:** choose **off, low, medium or high** in the chat menu or in your app's "reasoning effort". Off is the fastest. High is best for hard questions. - **Pictures:** say yes to "Images?" in setup. Then click **Picture** in the chat, or attach pictures in your app.
<p align="center"><a href="https://buymeacoffee.com/strataengine"><img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" height="50"></a><br> <sub>Strata is free. If it runs well on your PC, a coffee keeps the work on it going.</sub></p> ## What you need | | | | --- | --- | | **Graphics card** | **NVIDIA** GeForce RTX 20, 30, 40 or 50 series, or **AMD** Radeon RX 7900 XT / XTX, RX 7800 XT / 7700 XT, RX 9060 XT, RX 9070 / 9070 XT, Radeon AI PRO R9700 or RX 6800 / 6900 series. It needs **12 GB of VRAM or more**. |
| **RAM** | 32 GB or more. Your RAM decides [which model](#which-model-should-i-pick) fits. 64 GB runs every size. | | **Disk** | About 80 GB free. Use an SSD if you can: the first start is much faster. | | **System** | Windows 10 / 11 or Linux, and a current graphics driver from NVIDIA or AMD. | The installer sets up everything else. Two or three cards can share the model ([multi-GPU](docs/MULTI_GPU.md)). Experimental, written and tested by community members on their own machines:
download stops, run it again: it continues where it left off. Your browser opens the Strata app at `http://127.0.0.1:8080`. > **While the model starts, your PC can be slow or stop responding for 1-3 minutes** (longest the first time). > Strata loads 35-55 GB into your RAM and locks part of it for the graphics card. This is normal. Wait, and don't > close the window. The window shows what Strata is doing. **Next time**, run `START-HERE.bat` (or `./setup.sh`) again. It starts right away and downloads nothing twice. Close
AMD cards read pictures on Linux through the processor; on Windows they can't yet. - **From your phone or another PC:** `START-HERE.bat --setup --host 0.0.0.0 --api-key <secret>`. Always set a key. - **One request at a time:** by default Strata answers one request, and the others wait. To answer several at once, set `"parallel": 2` ([BATCHING.md](docs/BATCHING.md)). On a 12 GB card this makes each answer slower. - **Long prompts:** Strata reads the first message of a chat in full, about 1 minute per 30,000 tokens. Follow-up messages start in seconds.
- browser-use/jev-ultrafastPython · 22,856 Stars
网页自动化
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
Fastest and cheapest web agent
编报时的 AI 中文解读
Jev Ultrafast 是一个网页自动化代理，通过动态索引操作空间，让用户用自然语言目标驱动浏览器执行点击、输入等操作，适用于快速完成网页任务。
编报时的 README 依据
cp .env.example .env # Add TYPESAFE_API_KEY and TEXT_MODEL_API_KEY. uv run jev ``` Open **http://127.0.0.1:8766** and click **Start demo → Run automatically**. The inspector shows numbered elements, operation probabilities, target probabilities, and executed actions. **Choose next** pauses before execution. Chrome connects through [Browser Harness](https://github.com/browser-use/browser-harness), installed by `uv sync`. Run `uv run browser-harness --doctor` if it needs connecting. Allow remote debugging in Chrome when prompted.
`TEXT_MODEL_API_KEY` is an OpenRouter key in the example configuration. The current demo uses `inception/mercury-2.5` with reasoning disabled. Gemini, GLM, and DeepSeek can also use the OpenAI-compatible text helper; configure the appropriate model, endpoint, and reasoning setting. ## Use the library ```python from jev_ultrafast import Agent with Agent( "https://www.google.com/travel/flights?hl=en", "Find one-way flights from Zurich to London on September 20, 2026, " "for one adult in economy. Stop when matching flight options are visible.", ) as agent:
A `DONE` choice still requires independent outcome verification. The DOM reader handles common HTML and ARIA controls, not the full accessible-name specification. Shadow roots, frames, canvas, uploads, pop-up tabs, nested scrolling, and arbitrary keyboard widgets remain outside this MVP. Owned tabs share the existing Chrome profile. ## Development ```bash uv run ruff check . uv run pytest node --check jev_ultrafast/static/app.js node --check jev_ultrafast/snapshot.js uv build ```
- robbietilton/CompositorSwift · 16,794 Stars
图像编辑、PSD编辑、照片调色
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
The Photoshop alternative for Mac
编报时的 AI 中文解读
Compositor 是一款面向 Mac 用户的免费开源图像编辑器，旨在作为 Adobe Photoshop 的替代品，专注于合成和后期处理工作流，提供图层、蒙版、调整、滤镜等完整功能。
编报时的 README 依据
Get Compositor from [robbietilton.com/compositor](https://robbietilton.com/compositor), or download the latest release directly from [GitHub Releases](https://github.com/robbietilton/Compositor/releases/latest). ### Homebrew ```sh brew install --cask robbietilton-compositor ``` ## Features ### Layers - Layers and folders, with opacity and Photoshop's full set of blend modes in its order — a folder's opacity dims everything inside it
Compositor is English only for now, and translation pull requests aren't being accepted. It's maintained by one person, every new string would need translating from then on, and translations in languages I don't read can't be reviewed. I'll revisit this once the app settles; until then, please don't open localization PRs. ## Building Open `Compositor.xcodeproj` and run the **Compositor** scheme. ## Releasing `scripts/release.sh` builds a Release version, signs it with Developer ID, notarizes and staples it, and packages it into `dist/Compositor-<version>.dmg`.
- AI agents and scripts can build and edit projects directly: a `.comp` is a folder of PNG layers and a manifest, and an open project updates live as it's written. See [Writing Compositor projects](docs/writing-comp-files.md) ## Requirements - macOS 26.0 or later on a Mac with Apple silicon - Xcode 26 or later (to build from source) ## Translations
- Import JPEG, PNG, HEIC, TIFF, SVG, camera RAW (with a develop step first) and Photoshop PSD and PSB (8-bit RGB; not CMYK). Photoshop folders, masks, blend modes, fill rectangles/ellipses, and simple horizontal text stay editable; other vectors and vertical text become pixels. A conversion report is shown before anything is applied. - Large documents: the memory budget scales with your Mac, and a Photoshop file too big to open has its layers cropped to the canvas instead
- openai/mathLean · 14,718 Stars
数学研究资料、形式化证明
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
编报时的 AI 中文解读
该仓库包含由内部OpenAI模型生成的数学手稿及配套证明工件，涵盖多个数学领域，并附有Lean形式化证明。适合数学研究者、形式化验证开发者及对AI数学能力感兴趣的人士查阅。
编报时的 README 依据
Some of the unformalized results could have issues. We will endeavor to fix any such issues quickly. We are also exploring community-hosted repositories for these materials. ## Navigating the collection The current catalogue contains 719 manuscripts organized into 372 families. A family groups related papers, which may include a principal result, companion arguments, consequences, or alternative proofs. Each family is classified by mathematical discipline. - Start with the [overview](overview.pdf) for descriptions of the families.
- Use the [manuscript map](CONTENTS.md) to find individual papers and their supporting materials. - The [`preprints/`](preprints/) directory contains PDFs, source files, and manuscript-specific citation and build instructions. - The [Lean library](lean/README.md) and [formalization catalogue](lean/formalization.yaml) describe the available formal proofs, their associated papers, and verification configurations. See the [Comparator instructions](lean/ComparatorChallenges/README.md) for additional checking instructions. The repository has ~42% top-line results formalized.
- cdyforever/how-to-live-betterHTML · 11,695 Stars
在线阅读
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
《高性价比人生指南》全书 528 条的在线单页阅读版：手机可读、可搜索、零依赖、支持离线
编报时的 AI 中文解读
将开源书《高性价比人生指南》的全部内容渲染为单文件网页，供手机和电脑在线阅读与搜索，支持离线使用。
编报时的 README 依据
# 高性价比人生指南 · 在线阅读页 一个单文件 HTML 阅读页，把开源书《高性价比人生指南》的全部 32 节、528 条建议 渲染成可搜索的页面。手机上打开就能看，不用装任何东西。 **在线阅读：** https://cdyforever.github.io/how-to-live-better/ ## 这是什么 原书由 [eternity4719/HowToLiveBetter](https://github.com/eternity4719/HowToLiveBetter) 维护（Unlicense，公有领域），按「性价比」排序，每条写清花掉什么、换回什么、证据多硬， 只引期刊论文和官方文件。 本仓库不是原书的 fork，也不改动原书内容。它只做一件事：把原书正文渲染成一个 **自包含的单文件网页**，方便在手机和别人的电脑上直接打开。 ## 页面特性 - **零外部资源**：没有 CDN、没有字体外链、不 fetch 任何数据，断网也能读 - 逐条卡片：建议标题、证据等级（A/B/C）、性价比档、成本标签 - **「说人话」高亮块**——原书里这一栏专门把统计量翻成日常说法，页面上做得最显眼 - 全站关键词搜索（结果高亮）、只看 A 级、只看说人话 - 左侧目录带性价比色点，滚动联动 - 手机端：搜索框独占一行、顶栏向下滚动后自动收成一行、回到顶部按钮
- 明暗主题切换 - 支持直接打印 ## 本地使用 不用构建，双击 `index.html` 即可。 ## 重新生成 ```bash # 需要先有一份上游仓库的本地克隆 git clone --depth 1 https://github.com/eternity4719/HowToLiveBetter.git /tmp/upstream # 全部 32 节 python build.py all -o index.html --repo /tmp/upstream # 只做某几节 python build.py 1 2 16 --repo /tmp/upstream ``` 源目录也可以用环境变量 `HLTB_REPO` 指定。 统计口径（条目数、A/B/C 分级、性价比三档）与上游的 `tools/sync-stats.ps1` 和 `index.html` 保持一致，生成结果可与上游徽章逐项对照。 ## 自动更新 `.github/workflows/rebuild.yml` 每天 06:00（北京时间）拉取上游重新生成， 内容有变化才提交。也可以去 Actions 页面手动触发。 ## 授权 原书内容为 Unlicense（公有领域）。本仓库的构建脚本同样不作任何权利保留。
- Ebony-Vinyl/dsh-our-free-modelJavaScript · 14,736 Stars
模型服务接入、用量统计、模型接口转发
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
在 dsh 里装上这个插件即可，无需登录、注册或填 API Key，就能使用包括 DeepSeek V4.1 Flash、Kimi K3 在内的前沿模型——完全免费，不限量。 All you do is install this plugin in dsh: no login, no sign-up, no API key — the frontier models are just there, DeepSeek V4.1 Flash and Kimi K3 among them. Completely free, with no usage cap.
编报时的 AI 中文解读
为 dsh 宿主提供免费模型接入的插件，无需登录或 API Key 即可使用 DeepSeek、Kimi 等前沿模型，并附带用量看板、本地转发和公告推送等功能。
编报时的 README 依据
- **插件升级**——当前/最新版本、检查更新、一键升级（含进度与失败原因）、最近一次升级历史、热重载按钮与文件监视开关。 - **首次启动公告**——分 5 页（前言 / 模型清单 / 使用步骤 / 功能介绍 / 公告与升级），确认一次后不再弹出，除非文案版本号被提升。 ## 独立本地服务（开发中） 仓库现在提供两个入口：下文安装的是 **DSH 插件**；`packages/standalone` 是 **无需安装 DSH 的独立本地服务**。从完整源码仓库运行： ```powershell npm run start:standalone ``` 独立服务默认使用 `~/.our-free-model` 和端口 `18900`，拥有自己的 API Key， 不读取插件数据。支持匿名/Kilo、EAC 和原有十三个账号渠道，提供 Chat Completions、Responses API，以及独立网页管理（模型、统计、渠道、EAC、设置、API Key）。 启动后打开终端打印的一次性管理链接即可使用；账号在独立页面登录，也可手动导入插件账号备份。 启动、参数与验证见 [独立服务说明](packages/standalone/README.md)。 独立服务的安装制品与 Release 尚未发布。 ## 安装 **命令行（纯 dsh web）** ```
dsh plugin --profile web add /绝对路径/dsh-our-free-model ``` 安装完成后重启应用一次。`--profile` 填写实际使用的 profile 名称。 **DSHEAC AIO / 桌面端：请先阅读本节** 桌面端在启动 web 服务之前会执行一道 profile 闸门。其扫描器仅放行 dsh 自身在 .dsh-module-fallback 下生成的链接；profile 目录树中出现任何其它符号链接或目录 junction，应用将拒绝启动并报错： ``` PROFILE_UPGRADE_REQUIRED: offline dependency migration is not yet available ``` 因此桌面端不要使用 `link:` 依赖安装，也不要创建 junction。请使用应用内的插件管理器， 或放置一个真实目录。 以真实目录手工安装时，在 /profiles// 下完成三项操作： 1. 将发布文件复制至 node_modules/dsh-our-free-model/
<div align="center"> <img src="icon.svg" alt="Our Free Model — DeepSeek Harness 免费模型插件" width="120"> # dsh-our-free-model **简体中文** | [English](README_EN.md) <img alt="许可证" src="https://img.shields.io/badge/license-MIT-263146?style=flat-square"> <img alt="零依赖" src="https://img.shields.io/badge/dependencies-zero-4b6fff?style=flat-square"> <img alt="无构建步骤" src="https://img.shields.io/badge/build%20step-none-7da1de?style=flat-square"> <img alt="适配内核" src="https://img.shields.io/badge/dsh-0.1.5--0.1.7--rc.2-2f6f4f?style=flat-square">
- **EAC 渠道（桌面端专属）**——在 DeepSeek Harness 桌面端与 DSHEAC AIO 桌面端中自动解锁一条协付通道，模型以 EAC 前缀显示（如 EAC DeepSeek V4.1 Flash）；凭据加密密封，由宿主指纹闸门把守；该渠道在服务器侧校验 GitHub 授权（登录并 star 本仓库）后才放行对话；在命令行及其它宿主中该通道完全不存在。详见「EAC 渠道」。 - **Kilo 渠道（免密免费池）**——内置 Kilo AI 公共网关的免费模型池（`isFree` 清单实时拉取，含 `kilo-auto/free` 自动路由），无需任何账号或 Key；模型卡带 Kilo 徽章。思考强度与 EAC 渠道同款：模型自身的档位菜单（Off / Low / Medium / High，默认 High），经网关统一的 `reasoning` 参数真实下发——Off 已逐家族实测将思考归零（nemotron、ling、dots、poolside、apodex、cohere）；stepfun 与 liquid 端点强制思考（对关闭请求返回 400）、两个自动路由不透传关闭，这些模型的菜单不含 Off 档。该池由上游免费提供，上游会在其模型卡中声明 prompt 可能被记录用于改进服务——请勿发送敏感内容，详见「免责声明」。
缺少某段 usage 时，该值不等于整轮完整 token 总量。思考检查点不写入统计文件。 升级前的旧统计仅包含物理请求记录，迁移后展示的回合数与失败数为基于这些记录的推算值， 看板会明确标注；升级后的回合按最终结果精确记录。 请求数、回合数与 Token 总量为累计值；速度、首帧延迟与热力图使用本机保留的历史窗口。 已在一次真实 MiMo V2.6 Flash · Deep 请求中恢复成功；此前一次恢复曾在 480 秒 截止时失败。这不代表全部模型均已实测，也不保证每个长思考回合都能生成答案。 下一节列出的 v1.3.1 记录描述的是恢复功能引入前的行为。 ## 验收情况 在 Windows 环境下针对真实上游实测。本节按轮次记录，并注明每项的验证方式——仅标注 「真实上游」与「实机点击」的条目才对应用户在界面上可见的行为。 ### Issue #12：验收状态 恢复实现、离线回归与真实上游结果单独记录在 docs/issue-12-recovery.md。合并前 review 修复了一处告警 重复累积的缺陷并补充回归用例，recovery 99/99、truncation 35/35、fingerprint 新增 11 项与 typecheck 均已通过；其中 12 项使用真实插件与本地 HTTP
- storytold/pdfcraftRust · 13,003 Stars
PDF处理
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
PdfCraft: An open-source PDF reader and editor written by scratch in pure Rust
编报时的 AI 中文解读
PdfCraft 是一款用纯 Rust 编写的开源 PDF 阅读器和编辑器，支持阅读、组织、合并、拆分和保护 PDF 文件，可在桌面和网页端运行。
编报时的 README 依据
## Combine and split without losing a thing **Combine files** merges any number of PDFs into one. Each file gets a bookmark, with its own bookmarks nested underneath. **Extract** copies the pages you select into a new document. **Split** divides a document every *n* pages, or before the pages you choose. Nothing quietly disappears along the way: - links and named destinations are rewired to the copied pages; - form fields stay interactive; - layers keep their on/off defaults; - attachments come along. Every page of a combined document renders pixel-identical to its source. ```sh
- **Scriptable.** The `pdfcraft-cli` tool (see [Built for agents, too](#built-for-agents-too)) covers inspecting, rendering, extracting text, editing, combining, extracting pages and splitting. Robustness sweeps run on the same engine as the app. ```sh pdfcraft-cli info form.pdf # structure as JSON pdfcraft-cli text paper.pdf --page 3 # reading-order text pdfcraft-cli edit in.pdf --rotate 1,2:90 --delete 5 --title "Q3" --out out.pdf ``` --- ## Built for agents, too
- **Native on macOS, Windows, Linux and FreeBSD**, and **in the browser** through WebAssembly, from the same Rust codebase. Windows builds come for x64, x86 and ARM64 (Windows on ARM, no emulation); every ARM64 change is tested on ARM64 hardware in CI. - **Private by design.** Documents never leave your machine. There's no account, no telemetry and no cloud processing. - **Engine first.** Parsing, rendering and editing live in reusable library crates. The interface is one swappable layer on top.
PdfCraft is young and moving fast. The aim is a workbench where you can view, organize, annotate, fill, sign and edit PDFs, at parity with Acrobat Pro. **Where it stands (October 2026), honestly:** about half of Acrobat Pro's offline features are in (88% of the must-haves), but that is roughly a third of the work, because the hardest parts are still ahead.
- **Thin or missing:** reliable editing of existing text (especially CJK), OCR beyond Latin script, Office import/export, signature timestamps and long-term validation, PDF/A/X/UA preflight, XFA forms and localization. - **Hardening:** fuzzing still turns up crashes and hangs on hostile files; each one is fixed with a regression test. Quality has not yet been compared with Acrobat side by side. **Next, in order:** our own renderer, hardening and a fidelity harness against Acrobat, editing existing content, then the Pro workflows (signatures, OCR, Office, preflight, XFA) and 1.0 polish.
- storytold/filmcraftRust · 12,728 Stars
视频剪辑、视频调色、音频编辑
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
FilmCraft: An open-source, nonlinear video editor written by scratch in pure Rust
编报时的 AI 中文解读
FilmCraft 是一款用纯 Rust 编写的开源非线性视频编辑器，旨在重新实现 Adobe Premiere Pro 的工作流程，提供剪辑、调色、音频处理、字幕和导出等功能，适用于熟悉 Premiere 的专业剪辑师。
编报时的 README 依据
<p align="center"> <a href="https://getartcraft.com/"> <picture> <source media="(prefers-color-scheme: dark)" srcset="docs/brand/artcraft-logo-white.svg"> <img alt="ArtCraft" src="docs/brand/artcraft-logo.svg" width="200"> </picture> </a> </p> <h1 align="center">FilmCraft</h1> <p align="center"> <b>Video editing, color and sound; an open-source, clean-room reimplementation of Adobe Premiere Pro, rebuilt in pure Rust.</b> </p> <p align="center">
<a href="#documentation">Docs</a> · <a href="#status">Status</a> </p> <br> FilmCraft is a non-linear editor for people who know Premiere: the same panels, workspaces, tools and shortcuts, so your hands already know where everything is. Underneath, it is new from the bitstream up. The H.264, HEVC, ProRes and AAC codecs are our own, written in Rust from the public specifications. A GPU compositor works in linear light. Frame math runs on exact integer time, so edits never drift. And every action in the app is a command that an AI agent can drive as precisely as you can. <br> ## Edit
- **Swappable UI.** The interface is one crate (`ui-egui`) over the engine, so a different front end can replace it without touching editing logic. <br> ## Get started ```sh cargo run --release -p filmcraft # the desktop app, with a demo project cargo run --release -p filmcraft -- --control 9876 # plus the JSON-lines control server cargo run --release -p filmcraft-cli -- commands # list every engine command cargo run --release -p filmcraft-cli -- mcp # MCP server (headless) ```
"params": {"clip": 87, "effect": "lumetri", "param": "curve_luma", "value": [[0,0],[0.25,0.21],[0.75,0.82],[1,1]]}}} {"method": "ui.screenshot", "params": {"path": "frame.png"}} ``` The trailer and the grades in these screenshots were built exactly this way, by an agent driving the running app. <br> ## Everywhere - **Native** on macOS, Windows and Linux, with a native macOS menu bar. - **The web:** the same engine and UI run in the browser via WebAssembly (`apps/filmcraft-web`, see [docs/web.md](docs/web.md)); every release ships it as `filmcraft-web-<version>.zip`.
An open-source, clean-room take on the Adobe Premiere Pro workflow: native on macOS, Windows and Linux, and in the browser via WebAssembly.<br> By the ArtCraft team. </p> <p align="center"> <a href="#license-and-credits"><img alt="License: MIT OR Apache-2.0" src="https://img.shields.io/badge/license-MIT%20%7C%20Apache--2.0-8b5cf6"></a> <img alt="Written in pure Rust" src="https://img.shields.io/badge/pure-Rust-6a3fd6?logo=rust&logoColor=white"> <img alt="Runs on macOS, Windows and Linux" src="https://img.shields.io/badge/runs%20on-macOS%20%7C%20Windows%20%7C%20Linux-8b5cf6">
- **Speed on big footage.** Hardware decoding works on macOS and Windows; Linux has no hardware path yet. Blend modes and the most common effects run on the GPU, but Lumetri, keys and export rendering still run on the CPU; H.264 encoding can use the hardware encoder on macOS and on Windows with an NVIDIA GPU (opt-in, Export ▸ Hardware encoding), and H.265 export is hardware-only there ([#30](https://github.com/storytold/filmcraft/issues/30)). - **No plugins.** No VST3 / Audio Units or OpenFX hosting. - **Delivery codecs.** H.264 is our only delivery-codec export; no HEVC or AV1 export yet.
- **Real-world media and platforms.** Our decoders are bit-exact on conformance streams, but camera and phone files in the wild are less tested. Windows and Linux get far less testing than macOS. - **AI features.** Few so far; speech to text is optional and off by default. Bug reports with real footage are the most useful thing you can send us: [open an issue](https://github.com/storytold/filmcraft/issues) or tell us in [Discord](https://discord.gg/artcraft). ## Architecture A layered Cargo workspace: | Layer | Crates | |---|---|
- storytold/lightcraftRust · 12,131 Stars
照片管理、照片调色、RAW处理
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
LightCraft: An open-source non-destructive photo editor written by scratch in pure Rust
编报时的 AI 中文解读
LightCraft 是一款用纯 Rust 编写的开源非破坏性照片编辑器，提供照片库管理和 RAW 开发功能，可在桌面和浏览器中运行。
编报时的 README 依据
LightCraft is a complete darkroom in a single native app. Every adjustment is **non-destructive**, so your originals are never touched. Every slider renders through a **scene-referred, wide-gamut, 32-bit float pipeline**: highlights roll off like film, shadows open up without halos, and colour stays clean from capture to export. <table> <tr> <td width="50%" valign="top"> ### ☀️ Light **Exposure, Contrast, Highlights, Shadows, Whites, Blacks**, with edge-aware local tone mapping (a guided filter on
<td width="50%"><img src="docs/images/bw-split.jpg" alt="Split before and after view of sand dunes: color on the left, black and white with grain and vignette on the right"><br><sub>Split view: colour original on the left, B&W with Clarity, Vignette and Grain on the right. <i>Demo library.</i></sub></td> </tr> </table> <br> ## Organize everything A library that stays out of your way: **All Photos**, **Recently Added**, **Picks**, **By Date**, **Albums** nested in **Folders**, and **Recently Deleted**. Rate with <kbd>0</kbd>–<kbd>5</kbd>, flag with <kbd>P</kbd> / <kbd>X</kbd> /
JSON parameters. The UI, the keyboard, the CLI, a JSON-lines control channel and an **MCP server** all dispatch through the same entry point. An agent can cull a shoot, develop it, mask a sky and export it, and *see* the result. ```sh lightcraft --control 7980 ~/Pictures/trip ``` ```jsonc {"method": "engine.execute", "params": {"command": "photo.flag", "params": {"flag": "pick"}}} {"method": "engine.execute", "params": {"command": "develop.set", "params": {"values": {"light.highlights": -45, "light.shadows": 38}}}}
- **MCP server.** `lightcraft-cli mcp` exposes the command registry to Claude (or any MCP client), alongside helpers for import, query, develop, mask, render (returned as an image) and export. It runs headless, or attached to the running app with screenshots, clicks and gestures. See [docs/mcp.md](docs/mcp.md). ```sh cargo build --release -p lightcraft-cli claude mcp add lightcraft -- "$PWD/target/release/lightcraft-cli" mcp ~/Pictures/shoot # headless claude mcp add lightcraft-app -- "$PWD/target/release/lightcraft-cli" mcp --connect 127.0.0.1:7980 # live app ```
Native on macOS, Windows and Linux. In the browser via WebAssembly. Drivable end to end by AI agents over MCP. </p> <p align="center"> <img alt="Pure Rust" src="https://img.shields.io/badge/pure-Rust-f2a516?style=flat-square&logo=rust&logoColor=white"> <img alt="macOS, Windows, Linux and Web" src="https://img.shields.io/badge/macOS%20%C2%B7%20Windows%20%C2%B7%20Linux%20%C2%B7%20Web-8a5800?style=flat-square"> <img alt="MCP server included" src="https://img.shields.io/badge/MCP-ready-8a5800?style=flat-square">
override these defaults. Ordinary development commands keep their own settings. **Chinese and Japanese text** need the shared font repo, an optional build input (official releases always include it): ```sh git clone https://github.com/storytold/craft-fonts ../craft-fonts CRAFT_FONTS_DIR=../craft-fonts cargo run --release -p lightcraft ``` Without it LightCraft builds and runs the same, but Chinese and Japanese text have no glyphs. Fonts are never committed to this repo; see [craftrules `standards/fonts.md`](https://github.com/storytold/craftrules/blob/main/standards/fonts.md).
- **Background rendering.** A worker pool renders the loupe, before/after and every visible thumbnail off the UI thread: drafts during drags, full quality on release. - **Local-first.** No account, no cloud, no telemetry, no subscription. Your catalog is an append-only log of human-readable operations you can diff, back up or replay. <br> ## Feature status LightCraft is young and moving fast. **Where we honestly stand** (details in the [roadmap](ROADMAP.md#where-we-stand)): - **By feature count we're at ~79%** of Lightroom (core features 98%), tracked row by row in
[docs/parity.md](docs/parity.md). - **As a day-to-day Lightroom replacement we're nearer 60–70%.** It's great for JPEG/DNG and most Nikon / Sony / older-Canon raws on one machine. - **The biggest gaps:** - **camera colour calibration:** Sony, Nikon, Panasonic, Fujifilm and Canon CR3 raws have guarded estimates from their camera JPEGs, with built-in ILCE-7M4, X-H2S and X-T4 profiles; measured calibration is missing, and other raws or rejected fits retain a neutral matrix;
- **compressed Olympus raws and unsupported CR3 variants:** these use embedded JPEG previews when present. Fujifilm lossless/lossy compressed RAF now decodes sensor data; [verification and existing-library reload instructions](docs/raf-compression.md); - **AI masks and denoise:** subject and sky selection are classical heuristics; - **HDR, video and the Classic Print / Book / Map modules.** - **What's next:** see [where we're going](ROADMAP.md#where-were-going). | Area | Status | |---|---|
- yetone/magpieGo · 12,014 Stars
模型切换、模型接口转发、用量统计
Stars 统计更新于 2026/10/14 14:25:28
查看原始资料
作者描述
Every agent's model. One place. Codex on DeepSeek, Claude Code on Kimi, from the menu bar.
编报时的 AI 中文解读
magpie 是一个本地网关应用，让用户从菜单栏为各种 AI 编码代理（如 Claude Code、Codex、Gemini CLI）选择模型，并统一管理多个提供商和订阅账户，支持路由和配额管理。
编报时的 README 依据
- **Keep going when a quota runs out**: a routing group moves to the next account or model, and the agent never sees the error. It's for you if you use more than one agent, more than one provider, or more than one account. If you use one agent on its vendor's own plan and that's enough, you don't need it. Runs on macOS, Windows and Linux (menu bar app, window, TUI, web UI and CLI), in Docker, and on Termux. ## Quick start **1 · Install.** Download the app from **[usemagpie.ai](https://usemagpie.ai)**, or run: ```sh curl -fsSL https://usemagpie.ai/install.sh | sh ```
## Community Questions, ideas, or a model that won't show up? Tell us on **[Discord](https://discord.gg/vGSnD3ZKQF)**. That is where feedback goes. **This repository doesn't take pull requests.** Only maintainers can open them. If you'd like a fix or a feature, describe it on Discord and we'll build it. If magpie saved you from editing one more config file, **a ⭐ helps others find it.** <a href="https://star-history.com/#yetone/magpie&Date"> <picture> <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=yetone/magpie&type=Date&theme=dark">
- shihabal3amri/DiPlayKotlin · 10,397 Stars
CarPlay接入
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
Independent CarPlay receiver for compatible Android head units. Wired and wireless public preview.
编报时的 AI 中文解读
为兼容的比亚迪安卓车机提供CarPlay接收功能的独立应用，支持有线和无线连接，界面类似DiAuto。
编报时的 README 依据
# DiPlay **CarPlay for compatible BYD Android head units.** Wired and wireless, with the familiar DiAuto interface. Independent app: `com.shihab.diplay`. > **BYD support scope:** These projects focus on BYD cars. They may work on other brands, but other brands are unsupported and there are no plans to add support or fix brand-specific incompatibilities. [Download & website](https://shihabal3amri.github.io/DiPlay/) · [Release](https://github.com/shihabal3amri/DiPlay/releases/tag/v0.2.15) · [Report a problem](https://github.com/shihabal3amri/DiPlay/issues/new/choose)
with generated group credentials and unverified requested frequency; see [Android 9 Wi-Fi Direct](docs/ANDROID9_WIFI_DIRECT.md). Android 10+ verifies its negotiated group frequency. - Wired USB and wireless CarPlay with local authentication. - BYD HUD navigation with arrows, distance and street names on verified firmware. - Car hotspot support, improved audio buffering and saved receive diagnostics. - Automatic address discovery, fixed-channel Wi-Fi fallbacks and successful-configuration memory. - Icon/text size, resolution and frame rate; applying a display change reconnects CarPlay.
 ## 0.2.15 — public preview Install on the **car**, not the iPhone. No jailbreak, dongle, Mac, account or authentication server is required for use. Core CarPlay does not require ADB; optional dashboard, battery, wheel-speed and parked-video features do. Your head unit must permit APK installation. The APK supports Android 7.1+ (API 25); Android 7.1–8.1 support is new and not yet confirmed on a vehicle. Wireless supports Wi-Fi Direct, the car’s existing hotspot or Existing Wi-Fi / Same LAN. Android 7.1–9 Wi-Fi Direct uses a firmware-dependent legacy path
- Local diagnostic export. Reports are sent only if you choose to share them. - Separate installation alongside DiAuto. Run one projection app at a time. This is **not an Apple-certified product**. The APK bundles an experimental accessory identity recovered from public Carlinkit firmware, not a newly provisioned MFi identity for DiPlay. A bundled private key is extractable. Acceptance after future iOS updates, reliability across head units and suitability of that identity for general distribution are unresolved. This release invites community testing; it is not a guarantee of universal
- jaredpalmer/kevPython · 9,089 Stars
文本分类、文本评分、模型微调
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
Jev-like family of decision models built on top of Qwen3.5/3.8 you can train and run on your own
编报时的 AI 中文解读
Kev 是一系列基于 Qwen3.5/3.8 的小型决策模型，可自行训练和运行，用于对文本进行是非、多选和评分等判断，并输出校准概率。适用于需要自动化决策或路由的场景，如客服工单分类。
编报时的 README 依据
The ticket mentions a return, a late delivery and a billing problem, and the department probabilities say so. That's why Kev returns probabilities instead of a single label: your code can route the confident cases and send the rest to a person. ### Use It From Python If you already call Jev, point your client at Kev and keep the rest of your code. The TypeSafe SDK is included in `uv sync --extra serve`: ```python from typesafe_sdk import Choice, Noul, Score, TypeSafeClient client = TypeSafeClient( api_key="local", base_url="http://127.0.0.1:8009", model="kev-latest", )
You'll need Python 3.12 or 3.13 and [uv](https://docs.astral.sh/uv/). The repo's `.python-version` makes `uv sync` use 3.13; torch has no wheels for 3.14 yet. ```bash git clone https://github.com/jaredpalmer/kev.git && cd kev uv sync --extra serve uv run --extra serve python -m kev.serve --run jaredpalmer/kev-4b --port 8009 ``` This starts Kev-4B on your machine: CUDA or ROCm if you have a GPU, MLX on Apple Silicon. The first run downloads the adapter and the base model. `--run` also accepts a local checkpoint directory or a Hub revision like `jaredpalmer/kev-4b@qwen3`.
**Length.** Kev-0.8B, 4B and 9B trained mostly on states of up to 384 tokens, with longer ones in their document and skill fine-tunes (up to 7,552 tokens), Kev-27B on states of up to 32,768. The server accepts states of up to 65,536 tokens, and 8,192 more for each question, and refuses a longer one with a 422 instead of cutting it. How far past its training length each model stays accurate is the "Validated context" column in [Models](#models). For Kev-0.8B, 4B and 9B that is 8,192 tokens: at 16k the measurement on real contracts can no longer rule out a drop of more than 3 points, and at 32k
- storytold/vectorcraftRust · 8,596 Stars
矢量插图、AI辅助绘图
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
VectorCraft: An open-source, vector art tool written by scratch in pure Rust
编报时的 AI 中文解读
VectorCraft 是一个用纯 Rust 从头编写的开源矢量插图工具，旨在重新实现 Adobe Illustrator 的工作流程，适用于矢量艺术家和插画师。 它支持桌面端（macOS、Windows、Linux、FreeBSD）和浏览器（WebAssembly），并提供 MCP 服务器，使 AI 代理能够像人类一样操作绘图。
编报时的 README 依据
- **Fast.** Multithreaded SIMD rendering off the UI thread. 20,000 shapes render in about 27 ms at full retina resolution while the interface stays at 120 fps. - **Robust.** Exact curve booleans (no "cannot perform operation"), unlimited undo via structural sharing, and property-tested file round trips. - **Open.** A documented native format (`.vectorcraft`, JSON), first-class SVG, PDF (and PDF-compatible `.ai`) import and export, PNG/JPEG/WebP export, Export for Screens, and SVGZ, templates, GIF, TIFF and BMP on open.
VectorCraft is under active development. [**ROADMAP.md**](ROADMAP.md) covers what ships today, the milestones, and honest time-to-parity estimates. **Where we are (2026-10-07):** roughly 69–75% of Illustrator's features exist and work, and about 40–55% of "a power user can't tell the difference". Everyday vector illustration is close to usable: drawing and path tools, Pathfinder and Shape Builder, paint, gradients, appearance and transparency, type with styles, threading and Hebrew/Arabic bidirectional layout, and
<p align="center"><sub><b>Gradient Mesh</b>: shaded spheres</sub></p> </td> <td width="50%" valign="top"> <img src="docs/images/art-pathfinder.png" alt="Pathfinder: a purple gradient crescent moon and three orange stars on a peach background" width="100%"> <p align="center"><sub><b>Pathfinder</b>: a crescent and stars from exact booleans</sub></p> </td> </tr> </table> ## Why VectorCraft - **Familiar.** Illustrator's layout, tools, menus, panels and shortcuts: the Pen, Direct Selection, Pathfinder, Smart Guides, Appearance, Swatches, Layers and more. You already know how to use it.
- **Agent-native.** Every menu item, tool gesture, panel and dialog can be driven over a JSON control channel and an **MCP server**, so Claude and other agents can draw, edit and export the way a person does. - **Everywhere.** One codebase for the desktop apps and the same UI in the browser. ## Quick start ```sh cargo run --release -p vectorcraft # desktop app cargo run --release -p vectorcraft -- examples/dusk-poster.vectorcraft cargo run --release -p vectorcraft -- --control 7979 # + JSON control channel
cargo run --release -p vectorcraft-cli -- mcp # MCP server (stdio) cargo run --release -p vectorcraft-cli -- run --in examples/ribbons.vectorcraft --export out.pdf # headless batch cargo run --release -p vectorcraft-cli -- bench examples/neon-drive.vectorcraft # render timing cargo xtask bundle # dist/VectorCraft.app (macOS) cd apps/vectorcraft-web && trunk build --release # web build → dist/web cargo xtask ci # fmt, clippy, tests, layering, wasm, vendor names ```
Japanese fonts come from [storytold/craft-fonts](https://github.com/storytold/craft-fonts), an optional build input (releases always use it): `CRAFT_FONTS_DIR="$PWD/../craft-fonts" cargo run --release -p vectorcraft` (an absolute path). Without it, Japanese text uses the installed system fonts. See [`docs/development.md`](docs/development.md#fonts-craft-fonts-optional-build-input). On machines with two graphics processors, VectorCraft renders on the power-saving (integrated) one by default on
**New to VectorCraft?** Download it from the [VectorCraft page on getartcraft.com](https://getartcraft.com/apps/vectorcraft). That's the easiest way to install it. **Want a specific build or format?** On GitHub, the [latest release](https://github.com/storytold/vectorcraft/releases/latest) has every build listed below, and [all releases](https://github.com/storytold/vectorcraft/releases) has earlier versions and their notes. `<ver>` in the file names is the version number, and `SHA256SUMS.txt` lists a checksum for every file. ### Windows | Build | Installer | Portable | |---|---|---|
doesn't write Affinity files. [Scope and limits](crates/affinity/README.md). The interface speaks English, Japanese, Traditional and Simplified Chinese, Spanish, French, Italian, Russian and Ukrainian (and Czech and Brazilian Portuguese in the menus). The scores are self-assessed, so the [honest assessment](ROADMAP.md#honest-assessment-2026-10-05) explains how far to trust them. **What's missing:** - 3D and Materials; - the Photoshop-style raster effects (Effect Gallery); - CJK composition for vertical type (vertical type itself has initial support, with a Japanese interface);
- Variables and scripting; - an interaction-fidelity pass covering every tool's modifiers and small behaviours; - packaging for Windows and Linux. **Where we're going:** next is the interaction-fidelity pass alongside the raster-effects package, then 3D and advanced type, then hardening and packaging for 1.0. The prioritized list is in [Where we're lacking](ROADMAP.md#where-were-lacking-in-priority-order). **Workspace:** `crates/{geom, color, doc, pathops, text, effects, trace, brush, render, svg, pdf, eps, cad, metafile, format, tools, engine, ui-egui, mcp, testkit}`
- zai-org/ZCodeTypeScript · 7,794 Stars
AI辅助编程
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
Z.ai's coding agent harness. Powerful, intelligent, extensible.
编报时的 AI 中文解读
ZCode 是一个 AI 编程工作台，提供桌面应用、浏览器界面和终端 Agent，支持本地运行和远程连接，适合开发者进行 AI 辅助编程。
编报时的 README 依据
# ZCode <div align="center"> <img src="public/logo/icons/1024x1024.png" alt="ZCode" width="128" height="128" /> </div> <p align="center"> <a href="https://applink.feishu.cn/client/chat/chatter/add_by_link?link_token=47ag983c-8fcb-4d6d-814b-5395193a712c&qr_code=true">飞书社群</a> · <a href="https://discord.gg/z9aBcQXZQ3">Discord</a> </p> <p align="center"> 简体中文 | <a href="README.en.md">English</a> </p> ZCode 是 AI 编程工作台，提供桌面应用、浏览器界面和终端 Agent。本仓库包含客户端、后端服务、共享 UI，以及 Agent CLI 与运行时源码。 ## 更新 - 2026-9-23：更新至 ZCode v3.14.3 版本。 ## 初始化
该命令同时启动 Web 开发服务器（默认 `http://localhost:5173`）和后端（默认 `http://localhost:3030`）；浏览器访问前者。`/ws` 和一般 `/api` 请求代理到本地后端，`/api/v1/oauth/token` 单独代理到当前配置的产品服务。 Agent 源码修改后，执行 `pnpm --filter @zcode/cli... build` 并重启服务。需要验证完整发行包时，按下方“ZCode 命令行版”打包章节解压运行。 ### ZCode 命令行版 命令行发行包包含 TUI、Web 和 Agent，统一使用 `zcode` 启动：无参数进入 TUI；第一个参数为 `--web` 时启动 Web；其他参数交给现有 Agent CLI 处理。两种模式都在本机运行，无需 Electron。 ```bash # 默认进入终端交互界面 zcode # 启动 Web 界面 zcode --web # 指定项目和端口，不自动打开浏览器 zcode --web --workspace /path/to/project --port 3030 --no-open # 查看 CLI 或 Web 参数 zcode --help zcode --web --help ```
准备 Git、Node.js **24.14.0** 和 pnpm **10.33.2**，版本以 [mise.toml](mise.toml) 为准。以下开发和打包命令均在仓库根目录执行。 ```bash pnpm bootstrap ``` `pnpm bootstrap` 安装 workspace 依赖、准备桌面本地运行资源，再执行 `build:bootstrap`。 Agent CLI 与运行时源码位于 [apps/zcode-cli/](apps/zcode-cli/)，作为普通目录随本仓库一起克隆，无需单独拉取或初始化 Git submodule。 根据需要选择其他初始化或构建入口： | 命令 | 用途 | | ------------------------------ | ----------------------------------------------------------------- |
默认目标为 macOS arm64，默认输出目录为 `packages/desktop/dist/`。`--os` 支持 `mac`、`win`、`linux`，`--arch` 支持 `x64`、`arm64`；实际打包与签名需要目标平台对应的工具和配置。 安装：双击打开产物 DMG，将 ZCode 拖入"应用程序"。本地构建未签名，首次打开若被 macOS 拦截，执行： ```bash sudo xattr -rd com.apple.quarantine /Applications/ZCode.app ``` ### ZCode 命令行版 构建入口为 `pnpm build:zcode`。脚本会依次构建 CLI/TUI、后端和 Web，收集 TUI 的原生库、worker 与运行时依赖，再组装发行包；运行发行包仍需要 Node.js，版本以 `mise.toml` 为准。 打包前必须设置下载根地址 `ZCODE_DIST_BASE_URL`（可放在 `.env`、`.env.local` 或环境变量中），也可以通过 `--base-url` 传入。以下地址是占位示例，发布时替换为实际托管地址： ```bash pnpm build:zcode --base-url https://downloads.example.com/zcode/
# 已配置 ZCODE_DIST_BASE_URL 时 pnpm build:zcode # 仅重新组包，复用已有的 Agent、后端和 Web 构建产物 pnpm build:zcode --skip-build # 查看版本、输出目录等可选参数 pnpm build:zcode --help ``` 默认版本取根目录 `package.json`，输出目录为 `dist/zcode/`： - `releases/<version>/zcode-<version>.tar.gz`：运行包。 - `releases/<version>/sha256.txt`：校验摘要。 - `latest.json`、`install.sh`：版本索引和安装脚本。 完整目录可上传到配置的下载根地址。安装脚本从该地址下载运行包，默认安装到 `~/.zcode/runtime`，并在 `~/.local/bin` 创建 `zcode` 命令。安装目录可通过 `ZCODE_DIST_HOME` 修改，命令目录可通过 `ZCODE_DIST_BIN_DIR` 修改。 旧 Lite 用户需要改用上述构建命令、环境变量和新的安装脚本。新安装不会删除旧 Lite 目录，也不会迁移或删除已有会话数据。 本地调试打包产物时，可直接解压运行，无需上传或安装： ```bash
- jev-chat/jev-chat-jarvisKotlin · 7,721 Stars
聊天辅助、回复草拟
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
The chat decision assistant: before you reply, Jev reads the chat, judges intent and risk, and drafts replies you fill in with one tap. You press send. Android · Windows · macOS · iOS | 聊天辅助决策工具：先看懂对方再回复，发不发由你。
编报时的 AI 中文解读
jev-chat 是一个聊天辅助决策工具，在回复前先分析对方意图、判断风险并生成回复草稿，用户可一键填入消息框，是否发送由用户决定。适用于在 Android、Windows、macOS 和 iOS 上使用聊天应用的用户。
编报时的 README 依据
- **Works where you chat** — WhatsApp, QQ, Feishu and X on Android; chat windows on Windows and macOS; any app through the iOS keyboard - **You press send** — jev-chat only fills the input box and never sends on its own - **Your keys, no server of ours** — requests go only to the model provider you configure, with your own key - **Open source** — MIT licensed, five apps across Android, Windows, macOS and iOS ## Products | Platform | Product | How you use it | Status | | :--- | :--- | :--- | :--- |
| Android | [Jev for WhatsApp](../global/README.md) (global edition) | Overlay on WhatsApp chats in English | Preview v0.1.0 | | Android | [Jev Chat Assistant](../cn/README.en.md) (Chinese edition) | Overlay on QQ, Feishu, X and WhatsApp | v1.7 | | Windows | [Jev for Windows](https://github.com/jev-chat/jev-chat-windows) | Sits beside your chat window and reads it with screenshots and on-device OCR | Released | | macOS | [Jev for macOS](https://github.com/jev-chat/jev-chat-jarvis-mac) | Overlay that reads the chat on screen and judges with a local model | Released (Apple Silicon) |
| | Global edition | Chinese edition | | :--- | :--- | :--- | | For | Chats in English on WhatsApp | Chats in Chinese on QQ, Feishu, X and WhatsApp | | Download | [jev-whatsapp-v0.1.0-release.apk](https://github.com/jev-chat/jev-chat-jarvis/releases/download/global-v0.1.0/jev-whatsapp-v0.1.0-release.apk) | [jev-assistant-v1.7-release.apk](https://raw.githubusercontent.com/jev-chat/jev-chat-jarvis/main/cn/apk/jev-assistant-v1.7-release.apk) | | Guide | [Global edition README](../global/README.md) | [Chinese edition README](../cn/README.en.md) |
1. **Read.** Jev reads the conversation on your screen: through the accessibility service on Android, the chat window on desktop (screenshots with on-device OCR, or system text APIs), and the message you copy on iOS. It does not modify the chat app or log in to it. 2. **Judge first.** Before any reply is written, a judgment model works out what the other person really wants, how risky the moment is and the best way to respond. 3. **Draft, check, rank.** Jev drafts replies, checks them against that judgment and ranks them. Tap one to fill it into the message box. You press send.
| <img src="../global/docs/images/states/02-decide.png" width="230" alt="Global edition · analysis" /> | <img src="../global/docs/images/states/04-results.png" width="230" alt="Global edition · scored replies" /> | <img src="assets/images/overlay.png" width="230" alt="Chinese edition · overlay" /> | ## Download & Installation ### System Requirements - **Android**: Android 11 or later (the Chinese edition needs an ARM64 phone) - **Windows**: Windows 10 1903 or later, or Windows 11 - **macOS**: macOS 13 or later on Apple Silicon - **iOS**: build it yourself with Xcode ### Android Users
Download the latest version from the [Releases](https://github.com/jev-chat/jev-chat-jarvis-mac/releases) page. Apple Silicon only. See the [project README](https://github.com/jev-chat/jev-chat-jarvis-mac) for setup. ### iOS Users There is no App Store build yet. Build the keyboard from [source](https://github.com/jev-chat/jev-chat-jarvis-ios) with Xcode. ## How It Works
> **Use at your own risk:** Using Jev inside third-party apps such as QQ, Feishu, X, or WhatsApp may not comply with those apps' terms of service, and your account could be restricted or banned. Decide for yourself whether to use it. ## Community and Feedback - Problems with the global edition: [report them with the WhatsApp bug template](https://github.com/jev-chat/jev-chat-jarvis/issues/new?template=whatsapp-assistant-bug.md). - Anything else: [open an issue](https://github.com/jev-chat/jev-chat-jarvis/issues).
- Droid-Deck/DroidDeckKotlin · 7,716 Stars
游戏运行、Steam客户端
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
DroidDeck brings the SteamOS experience to Android
编报时的 AI 中文解读
DroidDeck 将 SteamOS 体验带到 Android 设备，在 Adreno 手持设备上运行 Valve 的 Steam 客户端 Big Picture 模式，并通过 ARM64 Proton 运行 Windows 游戏。
编报时的 README 依据
Use Android 9 or newer on a supported Adreno device (730 or newer, or 8xx). Adreno 6xx is experimental: DirectX 11 uses DXVK 2 and may run, and DirectX 12 games can still crash. Mali, Xclipse, PowerVR, and Adreno 710 are unsupported. No root is required. Allow about 3 GB for the runtime and 1.1 GB more for the desktop and emulators. Install the APK from [Releases](https://github.com/Droid-Deck/DroidDeck/releases), install the Linux runtime, then press **Play** and sign in. Steam downloads on first launch. Install **Desktop & apps** to use the desktop and emulators. The **Store** installs
- tamaratran/fast-jev-compactionTypeScript · 7,620 Stars
上下文压缩、聊天辅助
Stars 统计更新于 2026/10/14 14:25:29
查看原始资料
作者描述
Claude Code plugin that replaces the compaction summary with Jev decisions: every tool call and result is scored in one fast request, stale ones are dropped or truncated, everything kept stays verbatim.
编报时的 AI 中文解读
这是一个Claude Code插件，用于在上下文压缩时用Jev决策替代摘要，保留原始消息，同时也可作为npm库使用。适用于需要精确保留对话历史的Claude Code用户。
编报时的 README 依据
# fast-jev-compaction Claude Code plugin that replaces the compaction summary with Jev decisions: every tool call and result is scored in one fast request, stale ones are dropped or truncated, everything kept stays verbatim. Also usable as an npm library. ## What and why Most context compaction asks an LLM to summarize old turns. A summary is lossy: a file path, exact error, constraint, or command can disappear even when it matters later. This library never rewrites anything. It only deletes tool calls and tool results Jev says are no longer needed, and it asks Jev while
7. The message list is rebuilt: a message that loses all its content is removed, untouched messages are returned as the same objects, and no result is ever left without its call. Jev failures, malformed answers, a missing key, or a history that cannot be fitted throw; the caller (or the Claude Code hook) decides what to fall back to. ## Install and usage ```sh npm install fast-jev-compaction export TYPESAFE_API_KEY=... ``` ```ts import { compactMessages, reductionRatio, type Message } from 'fast-jev-compaction'; const transcript: Message[] = [
Then add this repository as a plugin marketplace and install the plugin, either from the shell or as slash commands inside a session: ```sh claude plugin marketplace add tamaratran/fast-jev-compaction claude plugin install fast-jev-compaction@fast-jev-compaction ``` The install prompts for the plugin options (API key, thresholds, `truncateHeadChars`, …); leave them at their defaults to use `TYPESAFE_API_KEY` from the environment. Restart Claude Code or run `/reload-plugins`. From then on `/compact` (and auto-compaction) goes through Jev: the toast reads
is a thin adapter that feeds `session.compact` transcripts through `src/` and falls back to Claude Code's built-in summary on errors or insufficient reduction. See [`hooks/README.md`](hooks/README.md) for configuration and the Claude Code 2.1.274 type reference. ### Install in Claude Code Function hooks are an early-access Claude Code feature (2.1.274+), so the opt-in flag must be set wherever Claude Code runs, e.g. in `~/.claude/settings.json`: ```json { "env": { "CLAUDE_CODE_ENABLE_FUNCTION_HOOKS": "1", "TYPESAFE_API_KEY": "<your key>" } } ```
| `maxStateTokens` | `25000` | Estimated token ceiling for the state | | `maxRequestTokens` | `30000` | Estimated ceiling for state plus one batch of questions | | `truncateHeadChars` | `300` | Characters of a dropped tool result retained before its note | `result.stats` reports message and character counts before and after, the per-reason decision counts, the state size in estimated tokens, which fitting stage was needed, and the number of requests. ## Limitations - Only tool calls and results are candidates; text messages are never removed
or shortened in the output (they are only abridged in the state Jev sees). - Token sizes are estimates from character counts, not a tokenizer. - Calibration is at the request level; a probability is not a proof that a result is safe to delete. The assistant can always re-run the tool. - The full state is repeated with every request, so a history near the state ceiling costs one request per handful of questions. ## Claude Code plugin The repository root is a Claude Code function-hook plugin: `hooks/fast-jev.ts`
- KKKKhazix/AIHOTTypeScript · 7,566 Stars
热点追踪、日报生成、资讯聚合
查看原始资料
作者描述
一个自己找热点、自己写日报的网站框架。把信源和精选标准换成你的，它就是你的行业热点站。
编报时的 AI 中文解读
AIHOT 是一个可定制的行业热点网站框架，能自动从信源采集资料，通过大模型筛选、评分、聚簇生成热点和日报，适合为特定行业构建热点站。
- rehan-remade/universal-modderPython · 7,464 Stars
游戏模组制作、逆向工程、资源生成
查看原始资料
作者描述
Point Claude at any game. Skills, tools and the fal MCP that let Claude Code mod almost any PC game you own: recon, reverse engineering, fal-generated art/3D/audio, in-game testing, showcase videos.
编报时的 AI 中文解读
为AI编码代理提供技能、工具和知识库，使其能够为PC游戏制作模组，包括逆向工程、资源生成、游戏内测试和展示视频制作。
- mizorewww/laya-mlxPython · 6,918 Stars
模型本地部署、模型接口服务
查看原始资料
作者描述
Native MLX runtime for Laya typed decision models — 7–14 ms short decisions on M3 Max. No text generation, PyTorch, or cloud API.
编报时的 AI 中文解读
在 Apple Silicon 上本地运行 Laya 类型化决策模型的 MLX 运行时，提供快速、低延迟的推理，无需云端 API。
- Mak5er/AirCardSwift · 6,560 Stars
钱包卡面美化、锁屏主题制作
查看原始资料
作者描述
Apple Wallet Card Skinner for iOS 18+ (No Jailbreak Required)
编报时的 AI 中文解读
AirCard 是一款 macOS 应用，用于自定义 Apple Wallet 卡面外观和锁屏密码键盘主题，支持 iOS 18 及以上版本，无需越狱。
- CopilotKit/OpenDotsTypeScript · 5,231 Stars
AI代理工作区、网页自动化、聊天辅助
查看原始资料
作者描述
Your always-on AI coworkers that move between text, calls, and Slack.
编报时的 AI 中文解读
OpenDots 是一个开源模板，用于构建持久化 AI 代理工作区，代理可在文本、通话和 Slack 中工作，每个代理拥有独立的计算机环境，支持网页和移动端。
- Louis-CFM/coucouSwift · 5,142 Stars
AI代理监控、权限审批、聊天辅助
查看原始资料
作者描述
A tiny friend in your Mac's notch and on your iPhone that keeps an eye on your AI coding agents: Claude Code, Codex, Cursor, Gemini CLI, Antigravity and more. Approve from the notch or your Lock Screen.
编报时的 AI 中文解读
Coucou 是一款跨平台桌面应用，在 Mac 的刘海、Windows/Linux 的屏幕顶部边缘以及 iPhone 上显示 AI 编程代理会话，支持权限审批、聊天、文件拖放等功能。
- yi1108/printfilmPython · 5,141 Stars
视频剪辑、图像编辑
查看原始资料
作者描述
PRINTFILM: AI short-video marketing and AI short-drama creation platform
编报时的 AI 中文解读
PRINTFILM 是一个AI短视频和AI漫剧创作平台，支持从主题或剧本生成分镜、图像、视频并合成成片，适合内容创作者和营销人员使用。
- CopilotKit/openmuseTypeScript · 4,947 Stars
网页自动化、任务管理、PDF处理
查看原始资料
作者描述
A personal agent with a browser, terminal, files, and work that keeps going built with CopilotKit and AG-UI.
编报时的 AI 中文解读
OpenMuse 是一个个人代理应用，集成了浏览器、终端、文件管理和持久化任务执行，支持 iOS、Android 和 Web 平台，适合需要自动化完成复杂任务的个人用户。
- yihui-dev/awesome-opus5-5-videos未标注语言 · 4,891 Stars
视频创作参考
查看原始资料
作者描述
A growing collection of viral videos made with Claude Opus 5.5 and the prompts behind them. Watch each original next to a live remake on Skillry. Updated regularly.
编报时的 AI 中文解读
这是一个收集用 Claude Opus 5.5 制作的病毒式视频及其提示词的合集，每个视频都附有创作者分享的提示词或原始帖子链接，适合想用 AI 生成动画视频的创作者参考。
- TheoLeeCJ/SemIf-OpenJevPython · 4,817 Stars
语义决策、模型本地部署
查看原始资料
作者描述
Semantic ifs from open models, on a 3090 at home. Independent; not affiliated with Jev or TypeSafe.
编报时的 AI 中文解读
SemIf 是一个独立研究项目，用开放模型在本地（如 RTX 3090）实现运行时定义的语义决策，直接读取类型化选项的概率，无需生成文本。适合需要快速、可审计决策的开发者或研究者。
- shader-effects-inc/shadersTypeScript · 4,789 Stars
WebGPU特效、可视化设计
查看原始资料
作者描述
WebGPU components for React, Vue, Svelte, Solid, JS & Framer
编报时的 AI 中文解读
Shaders 是一个提供 WebGPU 特效组件的库，支持 React、Vue、Svelte、Solid 和 JavaScript，包含 200 多个效果组件，并配有可视化设计编辑器和 CLI 工具。
- omlahore/RemoveMacAISwift · 4,638 Stars
系统精简、存储清理、隐私设置
查看原始资料
作者描述
Debloat macOS: turn off Apple Intelligence, analytics, ads and pop-ups. A native app and CLI, and every change can be undone.
编报时的 AI 中文解读
RemoveMacAI 是一款用于精简 macOS 系统的工具，可关闭 Apple Intelligence、分析、广告和弹窗，并清理存储空间。它提供原生应用和命令行两种方式，所有更改均可撤销。适用于希望减少系统干扰、释放磁盘空间的 Mac 用户。
- alchaincyf/huashu-art-motionJavaScript · 3,896 Stars
艺术动画生成、解说视频制作、动画拆解复刻
查看原始资料
作者描述
艺术动画skill：35种艺术风格、9种解说语法，用代码让画动起来。
编报时的 AI 中文解读
这是一个面向AI编码代理（coding agent）的Skill，用于通过代码生成艺术动画，提供35种艺术风格配方卡、9种解说语法、参数化片段和长卷穿越片骨架，适合制作解说视频动画、艺术风格动画和名画穿越短片。
- nykooi1/vibe-wisePython · 3,542 Stars
AI辅助编程、学习编程
查看原始资料
作者描述
A Claude Code / Codex plugin that helps you learn how to build while AI writes the code.
编报时的 AI 中文解读
VibeWise 是一个用于 Claude Code 和 Codex 的插件，旨在让用户在 AI 编写代码的同时学习编程。它通过要求用户先描述方法、检查权衡并解释不熟悉的概念，让用户保持对设计的控制。适用于希望在构建中学习的初学者、初级开发者和探索新栈的经验丰富的工程师。
- QingYunA/answer-me-with-htmlJavaScript · 3,324 Stars
HTML回答生成、图表生成、视频生成
查看原始资料
作者描述
Answer me with HTML — an agent skill that answers hard questions with a one-page HTML you can actually read. 让 AI Agent 用一页 HTML 回答复杂问题。
编报时的 AI 中文解读
这是一个Agent技能，让AI Agent用一页可读的HTML回答复杂问题，替代大段文本输出。适用于希望获得结构化、可视化答案的用户，可集成到Claude Code、Codex、Cursor等Agent中。
- mcncarl/jianying-headlessPython · 3,072 Stars
视频剪辑、视频导出、AI辅助剪辑
查看原始资料
作者描述
Private source preview: native Jianying drafts, isolated editing/export, and standalone Agent Skill.
编报时的 AI 中文解读
面向剪映专业版 macOS 的本地自动化工具，通过结构化剪辑计划生成可编辑草稿，在独立副本中修改多轨工程，并调用本机剪映引擎导出 MP4。适用于 AI 视频工作流的工程交接、批量草稿生成和 Agent 辅助剪辑。
- nullmoth/nvidia-macos-driverRust · 2,926 Stars
显卡驱动
查看原始资料
作者描述
Metal driver for NVIDIA GeForce cards on macOS Sequoia
编报时的 AI 中文解读
为macOS Sequoia上的NVIDIA显卡提供Metal驱动的项目，支持Turing及更新架构的显卡，适用于Intel Mac和OpenCore系统。
- Supermedo/bloodborne_pcC++ · 2,493 Stars
游戏启动、图形增强、在线联机
查看原始资料
作者描述
Bloodborne running natively on Windows 10/11: launcher in 13 languages, DLSS, FSR 3.1/4, unlocked FPS, cheats, mods, one-click updates.
编报时的 AI 中文解读
这是一个让《血源诅咒》游戏在Windows 10/11上原生运行的工具，提供启动器、图形增强、作弊、模组和在线联机功能。适合拥有该游戏解密文件的玩家使用。
- Jakeschincariol/replica-skillPython · 2,460 Stars
应用克隆、逆向工程、应用测试
查看原始资料
作者描述
Eleven free Claude skills that clone any app: reverse-engineer it, rebuild it, test it for bugs, then fix what its users hate. Free, MIT.
编报时的 AI 中文解读
提供11个Claude技能，用于克隆任何应用：逆向分析、重建、测试、修复用户痛点，并完成品牌、上线和部署。适合希望快速构建竞品应用的开发者或创业者。
其余 278 个项目见收录列表。