HN 发布:Onlook (YC W25)——面向设计师的开源开发者工具
Launch HN: Onlook (YC W25) – An open-source developer tool for designers

原始链接: https://github.com/onlook-dev/onlook/

Onlook 是一款面向开发者和设计师的开源(Apache 2.0)、本地优先的桌面设计工具。它让你在画布上可视化设计和原型制作真实应用,同时你的实际产品实时运行——不是模拟稿。将其指向本机上的任何 Git 仓库,它会运行你自己的开发服务器,因此你设计的就是真实产品。 它不自带 AI 模型,而是让你连接已有账号(Claude Code 或 ChatGPT/Codex),后者会编辑你仓库中的真实文件;可通过 MCP 连接器添加更多上下文。你可以在多个仓库和页面间处理多个任务,对比探索想法,并将成果导出为自包含的 HTML、PNG、JPEG 或 PDF。 它被定位为 Figma、V0、Replit Agent、Base 44 和 Webflow 的开源替代品。macOS 桌面应用现已可用,Windows 和 Linux 正在候补名单中;你也可以从源码构建,无需账号或云密钥。官方构建默认开启分析,但可禁用;源码构建会先询问。图标来自 Central Icons,其许可不延伸到 fork。

Onlook(YC W25)发布了一款开源桌面设计工具,它在真实代码库上叠加画布,让设计师使用自己的 Claude Code 或 Codex 订阅直接在运行中的代码里探索和编辑界面。更改通过 git worktree 完成,并以可审查的 PR 提交。画布并排显示多个真实浏览器页面,每个页面都运行实际应用并使用真实数据;注入的脚本让用户点击元素,通过内联聊天定位编辑。代理还获得设计专用工具(排列画框、生成迭代、建议)以及用于可复用 HTML 草图的“Artifacts”,这些草图与仓库分离。它支持任意框架,可在本地运行,并通过 MCP 连接 Figma。桌面应用在 Apache-2.0 许可下免费;付费云协作功能正在规划中。团队在早期 Web/Storybook 版本之后重建了产品,以回应最初桌面应用持续被使用的情况。早期反馈包括 onboarding/账户问题(已修复)、并行变体冲突,以及对开放模型、插件和团队评审功能的兴趣。
相关文章

原文

Craft websites, prototypes, and designs with AI on a canvas built for creative work. See your real app running live, shape it visually, and let your own agent write the code. An open-source alternative to Figma, V0, Replit Agent, Base 44, Webflow, etc.

  • 🎨 Built for design. Onlook is the open-source developer tool for designers. You see your real app running live on a canvas, explore ideas side by side, and shape the interface visually instead of from a terminal. Design apps, websites, posters, graphics, and more with your local fonts and assets.
  • 🤖 Bring your own agent. We don't ship our own model. Connect the Claude Code or ChatGPT/Codex account you already have, and that agent does the work inside your project. Plug in more tools through MCP connectors to bring all of the context you need to your creative workflow. Work on multiple design tasks at once across repos and pages.
  • 📁 Works on your real repo. Onlook opens any Git repository on your machine and runs your own dev server, so what you design is your actual product, not a mockup that someone has to rebuild later.
  • 🖥️ A desktop app, local-first. Onlook runs on your computer. Your code stays on your computer, and the source build needs no Onlook account, cloud keys, or .env file.
  • 📤 Share what you make. Export artifacts as self-contained HTML, PNG, JPEG, or PDF to hand off, review, or post.
  • 🧡 Open source. Apache 2.0. Read it, fork it, and customize it yourself (we'd love to hear how you customize it!)

What you can do with Onlook

🖥️ Download the Desktop App (Recommended) – The fastest way to get started with Onlook.

  • macOS (Apple Silicon · Intel x64) → Download
  • Windows – Coming soon. Join the Windows waitlist here and be emailed when we have updates.
  • Linux – No Linux artifact is published and we're still gauging demand for Linux. Join the Linux waitlist here and be emailed when we have updates.

🧑‍💻 Build from source – SOURCE_BUILD.md covers building and running Onlook on a Mac. No Onlook account or cloud keys needed.

  1. Open a project. Point Onlook at any Git repository on your computer, or import one. Your code stays where it is.
  2. Onlook runs your app. It starts your own dev server locally, the same way you would from a terminal.
  3. See it live on a canvas. Your running app appears in frames on an open canvas, so you can look at pages, sizes, and ideas side by side.
  4. Design with your agent. Describe what you want in chat. Onlook hands the request to the Claude Code or ChatGPT/Codex account you connected, and that agent edits the real files in your repo.
  5. Watch it update. Changes show up on the canvas as your dev server reloads, and Git keeps track of everything that changed.
  6. Share the result. Export artifacts as self-contained HTML, PNG, JPEG, or PDF.

Because Onlook runs your project with its own tooling, it works with all web frameworks or languages your dev server already supports.

  1. Start a new empty project by clicking the "New Draft" button in the upper left corner of the projects page.
  2. Use the input box in the lower right corner of the screen to write what you'd like to be designed in Onlook.
  3. Answer clarifying questions from the agent, and wait for it to build
  4. Click to select individual frames for context in follow-up prompts, or double-click on a specific element to use the inline chat and guide the AI to edit that specific element.
  5. Prompt more ideas on one or multiple designs until you're ready to share your work with a colleague.
  6. Select the frame you'd like to share, and ask the agent to convert it into an Artifact for exporting.
  7. Export the artifact as HTML/CSS or as a PNG using the export button in the upper right corner of the frame.
  8. Paste the exported artifact in Slack, email, or any messaging tool.

If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also browse open issues.

See the CONTRIBUTING.md for instructions and code of conduct.

To report a security issue, follow SECURITY.md instead of opening a public issue.

Looking for the previous version of Onlook?

  • Onlook v1 (web): the browser-based version lives on the v1 branch of this repository.
  • Onlook Desktop (Electron, original): the first desktop version is at onlook-dev/desktop.

Onlook collects product usage analytics to help us improve the app. In the official download they are on by default; a build from source asks first. You can turn them off in Settings, or set STUDIO_TELEMETRY=0 to run or build without them. See the privacy policy for details.

Onlook's source code is distributed under the Apache 2.0 License. See LICENSE for details.

The license covers source code, not every asset in this repository. Onlook's icons come from Central Icons by Iconists. They are licensed to Onlook for its own product, and that license does not extend to forks. Onlook's brand marks, DMG art, and sounds aren't covered either. See ASSETS.md for what's covered and what to replace if you fork, and NOTICE for third-party notices.

联系我们 contact @ memedata.com