Skip to content
Meet Orbi, your AI tester and developer

An AI agent that tests your product, finds the issues and fixes them.

Give Orbi your site's address. It explores every feature like a real user, writes the user stories and acceptance criteria, tests every journey in real browsers, files each issue on GitHub and fixes it with Claude Code. Then it tests again, until the experience is measurably better.

Runs on your Mac with your own Claude Code. Your code never leaves your laptop.

Exploring example.com…
0 of 9 tasks doneUX 61

    Works like a human tester and developer, without stopping

    Leave a laptop running with Orbi open. You can watch it work: its plan, every task it ticks off, every browser it opens, every issue it fixes.

    Map the product

    Orbi opens Chrome, explores every page and builds a feature map: what each type of user can do and where.

    Write user stories and acceptance criteria

    For example, "As a user, I can create an account" or "As a user, I can update my password", each with Given/When/Then criteria. Everything is saved for your project.

    Create test users for every persona

    Visitors, new users, admins, buyers and sellers each get their own test account and browser profile.

    Test every journey in real browsers

    Functional, end-to-end, UX, UI, visual, accessibility, data and performance checks in Chrome, Firefox and Safari (WebKit), on desktop and mobile.

    File every issue on GitHub

    Bugs, broken journeys, confusing UX, visual glitches, wrong data and weak AI prompts, each with steps to reproduce and screenshots.

    Fix with Claude Code, one by one

    Orbi runs Claude Code in your repo to fix each issue. It also fixes prompts when the problem is an AI answer. Then it commits and deploys.

    Test again and prove the improvement

    Each fix is checked in the browser. The UX score is measured before and after, so you see the improvement as a percentage.

    Repeat until nothing is left to improve

    Orbi keeps improving simplicity, ease of use, accuracy and presentation until you stop it.

    What Orbi checks

    Each scenario is recorded and reported, with screenshots, on your project's dashboard.

    Functionality, end to end

    Every user story is tested against its acceptance criteria, with real form entry, real sign-ups and real journeys.

    User experience

    Friction, confusing wording, dead ends, too many steps and missing feedback, rated journey by journey.

    UI and visual

    Overflow, overlap, broken images, clipped text and layout differences between browsers and screen sizes.

    Multi-browser

    Chrome, Firefox and Safari's engine (WebKit), on desktop and mobile, compared side by side.

    Data accuracy

    "undefined", NaN, impossible dates, totals that don't add up and wrong numbers on the screen.

    Accessibility and performance

    WCAG checks with axe-core, keyboard access, contrast, slow pages, console errors and failed requests.

    Download SelfFix Orbi for Mac

    Orbi lives in your menu bar. Click it, type a task such as "test example.com and auto-fix", and it gets to work. A floating window shows Orbi, its plan, and every task it has finished.

    Requires macOS 14 or later, Node.js 22+ and Claude Code. Orbi uses your own Claude Code subscription to plan, test and fix.

    Or install from Terminal:

    $ curl -fsSL https://selffix.org/install.sh | bash
    1. Create a project on selffix.org with your site's address and GitHub repo.
    2. Install Orbi and click Connect on the Connect page.
    3. Type a task in Orbi, or press Start in Orbi on the website.