From a reference library to design references for AI.

UIDiscovery is a library of 6,540 real websites, web apps and iOS apps, cut into 110,736 references you can browse, search and save. Designers use it to see how shipped products solve a problem. Agents query the same library through our API and MCP server.

A screenshot shows what a product looks like. It doesn't say why it works: how the layout is structured, what the type scale is, how the sections are paced, which pattern each part is solving.

A designer can read those decisions off an image. An AI agent mostly can't — so it copies the surface instead. That's why "make me a site like X" so often returns X.

We're turning each reference into structured design knowledge: layout, hierarchy, type, colour, components and the pattern it represents, in a form an agent can reason about. The goal is simple — an agent should study real references before it builds, the way a good designer does, and then make something original.

Describe what you're making — the product, the audience, the moment — and get back the references that matter for it, with the decisions behind them spelled out. For a designer, that's a head start on research. For an agent, it's the context it was missing.

The most useful references are often deep inside a product: onboarding, empty states, settings, checkout, the moment something goes wrong. We'll keep going further into real flows, motion and more platforms, so there's a reference for the screen you're actually designing.

• Real products only. References are shipped work, not mock-ups.
• Credit where it's due. Every reference names its product and links to its source. Owners can ask for removal — see Sources & rights.
• Study, then design. We build for originality, not cloning — for people and agents alike.