Your coding agent finally has eyes.
Claude writes the code, then runs it, taps through it and fixes what it finds. You stop being the one who checks.
For Mac. About a minute to install. No account. Open source.
- You ask once. Claude builds it and launches it.
- Then it looks at the real screen and taps, like you would.
- And it tells you what it found, after it’s fixed.
Right now, you’re your agent’s eyes.
Done! Please run the app
and tell me how it looks.

Copy it. Switch. Paste it.
Switch back.
$ git diff | less
…412 more lines
Resets in 3 hours.
- It writes blind. So you run it, look at it, screenshot it and explain what broke.
- Every check costs a Command-Tab and your train of thought. Then you do it again.
- Vibex puts it all in one window Claude can see and drive. You get your attention back.
The loopIt checks its own work.
- Claude builds and launches your app, on its own.
- It reads what’s actually on the screen, like a tester would.
- It fixes what’s wrong. You come back to review, not to debug.
New · Vibex DesignSee it before you build it.

- Describe an app or a site. Claude writes the design as plain JSON in your repo.
- It appears on a canvas, screen by screen. Nothing here is a mock-up.
- A wrong idea costs you a sentence. Not a week of code.
- Love it? Build This hands the design to Claude.
Made with Vibex DesignClaude designed every pixel of this.

- These are real renders from Vibex Design, not drawings of what it might do.
- The result is JSON in your repo, so Claude can build exactly this.
Claude, nativeAll of Claude Code. None of the terminal.
- Type
/for every command and skill, with what each one does. - Plan when it matters. Claude proposes, and nothing changes.
- Auto when it doesn’t. It works on its own and stops only for risky actions.
- It’s the real Claude Code, with your CLAUDE.md, skills and hooks. There’s nothing to migrate.
Parallel agentsTry three ideas. Keep the best one.

sheet-checkout
0 0 · Pay button still clipped

one-page-checkout
0 0 · Checked on the simulator

apple-pay-first
0 0 · Waiting for you
⌘Returnsends a message as its own agent, on its own branch, in its own folder.- They work side by side and can’t touch your project. A bad attempt costs you nothing.
- Review each one, merge the winner, throw the rest away.
SimulatorThe iPhone, inside your window.


- No Simulator window lost behind everything else. The device lives next to your code.
- Claude reads every element by name, so it taps what it means to, not a guess at some pixels.
- A small phone, a big phone and an iPad, all live. Run on All installs everywhere at once.
BrowserIt reads the console so you don’t have to.
Eagle Ridge
- Your dev server opens in a tab next to Claude, and the address bar shows the route you’re on.
- Claude clicks through pages, watches requests and reads every error for itself.
- Never paste a stack trace into a chat again.
ApprovalsWalk away. We’ll get you when it matters.
- Long job? Go make a coffee. You don’t have to watch Claude work.
- When it needs a yes, you get a dot on its tab, a count in the window and a notification.
- One click takes you there. One click puts it back to work. No session sits waiting for hours.
ReviewEvery change, before it counts.


- Side-by-side diffs, right beside the running app. No more guessing what the agent touched.
- Every run saves a screenshot. Drag between before and after, and paste both into your PR.
- Commit and push from the same window. You stay the one who decides.
Skills and MCPNew abilities, one click away.
- Browse skills and MCP servers from inside Vibex, without leaving your work.
- It asks for any API key, then installs for this project or for all of them.
- You can always see what’s connected, what needs a sign-in and what’s failing.
AI UsageNever hit a limit by surprise.
- Session and plan limits at a glance, for Claude Code and Codex.
- A nudge before you run out, not after, so a big refactor doesn’t die halfway.
- It’s all read from logs on your Mac. Nothing leaves your machine.
The small things add up too.
One button to run
iOS, macOS, Node, Swift, Rust or Go. Vibex works out how your project runs, and Claude can start it too.
Files, always current
A file tree that updates as Claude adds and deletes things, with read-only code, Markdown and images.
Quick open
⇧⌘P finds any file or command, so you never go hunting through folders.
Use your favorite agent
The terminal runs Codex, Gemini CLI or Aider. The tools are standard MCP, so any client can use them.
One window per project
Each project keeps its own tabs, layout and Claude, right where you left them.
Native, and quick
Swift and AppKit, not a browser in a box. It updates itself in the background.
Free. Not free for now.
Vibex is MIT licensed and built in the open. Read every line, fork it, ship your own. You only pay for the AI tools you already use.
- No account, no subscription, no trial
- No tracking. Usage numbers stay on your Mac
- Use it at work, change it, keep it
A star costs you nothing, and it’s how other people find this.
What it won’t do, so you don’t find out later
- It’s Mac only, and needs macOS 26.
- The simulator pane needs Xcode 26 with an iOS runtime.
- Claude Code is set up for you, but it isn’t included. You bring your own plan.
- Vibex isn’t notarized yet. The install command handles that; with the DMG, macOS asks you to allow it once.
Questions
Is it really free?
Yes. It’s open source under the MIT license, with no paid tier, no trial and no account. You only pay for the AI tools you already use, like Claude Code.
Do I have to change how I work?
No. It’s the real Claude Code, so your CLAUDE.md, skills, hooks and MCP servers carry over. Your repo stays as it is, and a full terminal is one tab away.
What do I need?
A Mac on macOS 26, Apple silicon or Intel, plus Claude Code. Xcode 26 with an iOS simulator runtime is only needed for the simulator. Setup checks each one and installs what it can.
Does it work with Codex or other agents?
Yes. The terminal runs any agent CLI, like Codex, Gemini CLI or Aider. Vibex’s tools are a standard MCP server on your Mac, so any MCP client can use them, and AI Usage reads Codex’s limits too. The native chat and Claude Code setup are for Claude.
What can it run?
iOS and macOS apps from Xcode or XcodeGen, web apps with a dev server (npm, pnpm, yarn or bun, so Vite, Next.js and the rest), and Swift, Rust and Go command-line projects. Anything else runs in the terminal like it always has.
Is my code or usage sent anywhere?
Vibex sends your code nowhere itself. Claude Code talks to Anthropic as it always does. Usage numbers come from local logs. Vibex contacts GitHub for updates, skills.sh and the MCP registry when you browse them, and a server’s own site to fetch its icon.
Why does macOS warn me when I first open it?
Vibex isn’t notarized yet. The install command handles it. With the DMG, open System Settings → Privacy & Security and choose Open Anyway.
How does it update?
Vibex checks GitHub releases in the background and offers the new version in its own window, or you can run the install command again.
Give your agent eyes.
Paste one line. A minute from now, Claude can see what it builds.
Free forever. No account. Or download the DMG.
- 1. Paste the commandIt checks the release, installs Vibex and links the
vibexcommand. - 2. Open a projectSetup shows what’s missing, with one button to fix each thing.
- 3. Ask Claude for somethingThen watch it run, look and fix, instead of asking you to.