Skip to content
Start free

The AI design workbench

Make it look like you meant it.

Bring references, a repository, a live site or screenshots. M1rror’s design agent turns them into a real design system, designs every screen in every state, and gets it ready to ship as code.

Sign up free and this becomes your first project. You see what the agent’s work costs before it starts.

Or start from an example

Start freeSee how it works

Free plan, no card needed.

Six modes, one project shell

Board
Collect references, screenshots and notes.
System
Tokens and components, read from your sources.
Screens
Every screen, every state, every size.
Code
Your repository, checked against the system.
Review
Pinned comments and approvals.
Ship
An approved release, ready to build.

How it works

From a reference to a pull request, in five moves.

Each step leaves something real you can open, change and approve: a reference, a system, a screen, a decision, a release.

  1. Board

    Bring what you have

    Drop in references and screenshots, connect a GitHub repository, or capture a live web app you own. Or just describe it.

  2. System

    Get a real design system

    Colours measured from your pictures, type matched to open fonts, components and tokens you can name, compare and change.

  3. Screens

    Design every screen and state

    Phone, tablet and desktop at real sizes, with loading, empty, error and denied states designed, not left for later.

  4. Review

    Review together

    Pin comments to the exact spot, @mention teammates, and send review links to people without an account.

  5. Ship

    Ship it as code

    Approve a release and hand your coding agent a scoped build plan. Pull requests in your connected repository are checked against the system.

What M1rror does

Everything between the idea and the pull request.

Design agent

A design agent that shows you the price first.

Ask for a change in plain words, or select something and ask about it. The agent plans the work, quotes it in usage units and waits for your go-ahead.

  • Small asks just happen. Anything up to your auto-approve limit (5 units by default, 0 to always ask) runs straight away.
  • Big ones wait for you. Approve, change the plan or cancel. You’re charged only what it used.
  • No silent overruns. If the work needs more than you approved, it stops at a safe point and asks.

System

A real design system, read from your references.

Give M1rror a concept, a screenshot, a mood board or your code. It drafts the tokens and components, and tells you where it saw each one.

  • Measured, not guessed. Colours come from the pixels; type is matched to open fonts and marked when approximate.
  • Contrast fixed for you. Text pairs below 4.5:1 are adjusted, with your original kept as an option.
  • Keep several systems. Name them, compare two side by side, and switch when you’ve decided.

Screens

Every screen at real sizes, in every state.

Screens are designed on phone, tablet and desktop frames at their true sizes, from your design system’s own components.

  • States, not just the happy path. Loading, empty, error, denied, session expired and narrow are part of the design.
  • Ask for a state. The agent designs one from your system and leaves it as a draft you accept or undo.

Review

Feedback pinned to the exact spot.

Comment on the element you mean, not on a screenshot of it. Threads stay attached to the screen, component or board item they’re about.

  • Mention and resolve. @mention a teammate, reply, and resolve the thread when it’s done.
  • Draw on it. Box, circle or arrow the part you mean. The agent reads what each comment points at and can propose the fix.
  • Review links for guests. People without an account can comment on one scope, with an expiry and an optional passcode.
  • Reviewers are free. Commenting and approving never needs a paid seat.

Code

Your code, held to the system.

Connect a GitHub repository. M1rror reads it without running it, links your components to the design system and shows where the code drifts.

  • Components linked to code. Each component shows the file and export it lives in, with its states.
  • A design check on every pull request. Off-theme colours, off-scale spacing, low contrast and hand-rebuilt components are reported on GitHub.

Trust

Your product, your money, your call.

An agent that works on your account should be easy to trust. Here is what M1rror does, and what it won’t do.

  • You approve the spend

    Agent work above your auto-approve limit waits for a quote you approve. The limit is 5 units unless an owner changes it, and 0 means always ask. You never pay more than you approved; unused units go back.

  • Private by default

    References, captures, repositories and screens belong to your workspace. Every request is scoped to it on the server, so another workspace can’t reach them.

  • Not used for training

    Your code, references and feedback are not used to train public models.

  • Read, not run

    Code scans and pull request checks read your files. They never run your repository’s code, and M1rror never pushes to your default branch or merges.

  • People decide

    The agent prepares changes as cards you approve and never gets more access than you. It can’t approve its own work, and it never touches payments, sign-in, access grants or deletion.

  • Accessible from the start

    Every text colour M1rror proposes is checked for 4.5:1 contrast and fixed if it falls short. Components get focus and disabled states. The workbench itself is tested with axe in light and dark.

How security and governance workPrivacy policy

Pricing

Start free. Pay when you ship more.

Reviewers never need a paid seat. Every plan includes the design system and review.

  • Explore

    $0free

    Try the loop on a small project.

    • 150 usage units a month
    • 2 projects
    • 2 editors included

  • Studio

    $49per month

    For a builder shipping a real product.

    • 1,000 usage units a month
    • 10 projects
    • 2 editors included
    • MCP server for your coding agent

  • Team

    $199per month

    For small teams and agencies with reviewers.

    • 5,000 usage units a month
    • 50 projects
    • 5 editors included
    • MCP server for your coding agent
Compare plansStart free

Pilot pricing: billing runs in Stripe test mode and prices may change.

FAQ

Questions, answered plainly.

Something else? Read how security works or compare plans.

Your next screen starts with a sentence.

Describe it, bring a reference or connect your code. The Explore plan is free and needs no card.

Start freeSee pricing