Skip to main content
株式会社オブライト
株式会社オブライト

Onlook 2.0 Explained

Free Open-Source Visual Editor for Claude Code and Codex: Install, Pricing, vs Figma and v0

Published 2026/11/2210 min read

Onlook 2.0: free Apache 2.0 Mac app that shows your real app live while Claude Code or ChatGPT/Codex edits the code. Install, usage, pricing, vs Figma/v0.

Onlook is a free, open-source (Apache 2.0) desktop app that shows your real app, running from your own Git repository, on an infinite canvas and lets Claude Code or ChatGPT/Codex rewrite the actual code as you shape the design. It shipped as "Onlook 2.0" (tag v2.0.2) on GitHub on November 19, 2026, and drew attention on Hacker News through a Launch HN (YC W25). The official download is macOS only (Apple Silicon and Intel) as of today; Windows is listed as coming soon and no Linux build is published. This article covers how it works, how to install it, the shortest path to using it, how it compares with Figma, v0 and Cursor, and the caveats, based on the official README, release notes and website.

Everything here reflects information as of November 22, 2026. The official site lists no version number or price sheet, and pricing for the hosted cloud version is not public, so I could not confirm it. Where something could not be verified, the text says so.

What Onlook is: a developer tool for designers

The README describes Onlook as "the developer tool for Designers": a free, local-first editor where designers craft experiences in code. The key idea is that the code of the actual product is the source of truth, not a separate design file. What you see on the canvas is not an image or a mockup but the real app running on your own dev server.

The README positions it as an open-source alternative to Figma, V0, Replit Agent, Base 44 and Webflow, among others. Onlook ships no model of its own, though. The README states it does not bundle a model. You connect the Claude Code or ChatGPT/Codex account you already have, and that agent edits files inside your repository.

The earlier Onlook (v1) was a browser-based editor for Next.js and Tailwind projects. Version 2.0 is a rebuild as an Electron desktop app. v1 remains on the v1 branch, and the first desktop version lives at onlook-dev/desktop.

What it can do

AreaCapabilities
Design on your real codebaseLive app on an infinite canvas; mobile, tablet and desktop frames side by side; works on a safe worktree copy so your folder is never touched; parallel projects as tabs
Design with your own agentConnect a Claude Code or ChatGPT/Codex plan; ask about or edit any element; the agent arranges variations, states and flows on the canvas; reference images on the canvas
Sketch and shareArtifacts for quick HTML sketches outside your repo; export to HTML, PDF, PNG and JPEG
Ship itImport a GitHub repo; start a new codebase from scratch; one-press pull requests from checks to merge

You can also use local fonts and assets, and the README says you can design apps, websites, posters and graphics. MCP connectors bring in outside tools and context. The official site names Next.js, React and Vite as supported frameworks and Storybook and shadcn/ui as supported tooling, so design-system components can be used directly. The README adds that it works with any web framework or language your dev server already supports.

How it works in six steps

- 1. Open a project: point Onlook at a Git repository on your machine, or import one. Your code stays where it is
- 2. Onlook runs your app: it starts your own dev server locally, as you would from a terminal
- 3. See it on a canvas: the running app appears in frames so you can compare pages, sizes and ideas
- 4. Design with your agent: describe changes in chat, and the connected Claude Code or ChatGPT/Codex agent edits the real files
- 5. Watch it update: the canvas refreshes as the dev server reloads, and Git tracks every change
- 6. Share: export Artifacts as self-contained HTML, PNG, JPEG or PDF

The stack is public too: Electron, React 19, Astryx (design system), React Flow, Zustand and CodeMirror on the desktop side; Git and node-pty to run your dev server and shells; Bun and Turborepo for tooling. Exports use Electron's capturePage for PNG/JPEG and printToPDF for PDF. Being an Electron app matters when you think about memory use and updates (for another Electron-based agent tool, see our T3 Code guide).

Install and pricing

- Official download (recommended): onlook.com/install. macOS on Apple Silicon and Intel x64
- Build from source: SOURCE_BUILD.md covers building on a Mac. A source build needs no Onlook account, cloud keys or .env file
- Windows: coming soon, with a waitlist
- Linux: no artifact published; the team is gauging demand, with a waitlist

Pricing: the README calls the app free and local-first, and the release notes say it is free and works with the Claude or ChatGPT subscription you already have. So the app costs nothing, but agent usage is billed through your own Claude or ChatGPT plan. The hosted cloud version on the website is contact-us or book-a-demo, with no prices listed.

License: the source code is Apache 2.0, but not every asset is covered. Central Icons icons, brand marks, DMG art and sounds are excluded, and ASSETS.md lists what to replace if you fork.

Telemetry: in the official download, usage analytics are on by default and can be turned off in Settings. A source build asks first. Setting STUDIO_TELEMETRY=0 disables it for running or building. If your organization restricts outbound data, check this before rolling it out.

How to use it: the shortest path

1. Download the macOS app from onlook.com/install, launch it and connect your Claude Code or ChatGPT/Codex account
2. Click "New Draft" in the upper left of the projects page to start an empty project (or open a Git repo / import from GitHub)
3. Type what you want designed in the input box at the lower right
4. Answer the agent's clarifying questions and wait for the build
5. Select a frame as context for follow-ups, or double-click an element to use the inline chat and change only that element
6. Keep layering ideas on one or several designs
7. Pick the frame to share and ask the agent to turn it into an Artifact
8. Export as HTML/CSS or PNG from the button at the upper right, then paste into Slack or email

For developers, the reassuring part is that Onlook works on a worktree copy instead of your original folder. A worktree is a standard Git feature, so you can discard it or merge it back as a normal branch. That lowers the risk of an agent damaging your main branch, and the work still ends in a one-press pull request for review.

Onlook vs Figma, v0, Cursor and Claude Design

Onlook is likely to be searched in "Figma alternative" or "v0 alternative" contexts. The README names Figma, V0, Replit Agent, Base 44 and Webflow; it does not compare itself directly with Cursor or Builder.io. The table below sets Onlook's documented design beside the generally known positioning of the other tools. Details and pricing of those products change, so check their official sources before deciding.

AspectOnlookDesign tools like FigmaAI generators like v0Coding agents like Cursor / Claude Code
Source of truthReal code in your local repoDesign file (implementation done separately)Generated code (mostly kept on the service)Real code in your repo
What you seeYour real app on an infinite canvasDesign framesPreview of the generated resultEditor or terminal (check the screen in a browser separately)
Where the AI comes fromBring your own (existing Claude Code / ChatGPT-Codex plan)Varies by productBuilt into the serviceVaries by product
Where it runsLocal desktop (Mac)Mostly cloud / browserMostly cloudLocal or cloud
Best forDesigners and visually minded engineersDesignersPeople who want a fast prototypeEngineers
LicenseApache 2.0 (some assets excluded)CommercialCommercialVaries by product

In short, Onlook does not build an agent. It puts a designer-friendly surface on top of existing agents. The coding agent writes the code; Onlook provides the visual way to say what to change and a canvas to compare several options.

For related choices, see Claude Design vs Figma vs Canva for AI-made design, Figma MCP catalog setup for feeding Figma data to agents, and Codex vs Claude Code vs Cursor vs Copilot for the agents themselves.

Pairing it with a design skill

Onlook lets you direct changes visually, but whether the generated UI looks generic depends on the connected agent and its instructions. A design skill for Claude Code or Cursor such as Hallmark is one way to keep the tone consistent while you compare options on the canvas. That combination is not documented by Onlook and I have not tested it.

Caveats before adopting

- Supported OS: the official download is macOS only. Windows and Linux developers currently depend on a source build, and the README says its instructions are for Mac
- Agent costs: the app is free, but your Claude or ChatGPT plan limits and charges still apply, and repeated large edits consume them
- Telemetry: on by default in the official build; disable it in Settings or with STUDIO_TELEMETRY=0
- Maturity: the 2.0 tag was created on November 17 and published on November 19 (latest release v2.0.2), so expect change
- Code quality depends on the agent: Onlook provides the visual workflow, but generated code still needs normal review
- Hosted pricing: not listed on the website, so I could not confirm it

Summary: who should try it

Onlook suits designers and front-end developers who already subscribe to Claude Code or ChatGPT/Codex and want to refine their own product's screens visually in real code, not in a mockup. The official site cites more than 27,000 GitHub stars and the repository showed about 27.7k when I checked, so attention is high, but 2.0 is days old. A sensible first step is to fix one screen on a throwaway branch or worktree and see whether it fits your team's review process.

What is Onlook?

A free, open-source (Apache 2.0) desktop app that shows your real app on an infinite canvas and lets a connected Claude Code or ChatGPT/Codex agent edit the actual files while you shape the design. Your repository's code, not a design file, is the source of truth.

Is Onlook free?

The app is free and the source is Apache 2.0, but the AI work runs on your own Claude Code or ChatGPT/Codex plan, whose usage and limits are billed separately. Pricing for the hosted cloud version is not listed on the website, and I could not confirm it.

Does Onlook run on Windows or Linux?

As of November 22, 2026 the official download is macOS only (Apple Silicon and Intel). Windows is coming soon and no Linux build is published; both have waitlists. The README's source-build instructions are also written for Mac.

Does Onlook include its own AI model?

No. The README says it does not ship its own model. You connect your existing Claude Code or ChatGPT/Codex account, and MCP connectors can add context from other tools.

Will it break my existing repository?

The README says Onlook works on a safe worktree copy and never touches your original folder. Changes are tracked in Git and can become a pull request in one action, though you should still review generated code as usual.

Does Onlook collect usage data?

In the official download, usage analytics are on by default and can be turned off in Settings. Setting STUDIO_TELEMETRY=0 disables them for running and building, and a source build asks before enabling them.

株式会社オブライト

Oflight Inc. is a Tokyo-based IT company providing software development, digital signage, web development, network systems, and IoT/AR/VR technology services.

Related free tools (no sign-up, instant results)

Feel free to contact us

Contact Us