Kage turns real product designs into prompts for coding agents

The free library gives Claude Code, Codex and Cursor visual references drawn from 160 designs across 88 products.

By · Published

Primary source: Kage

Why it matters

AI coding tools increasingly need structured visual context alongside code instructions. Kage packages real product references as agent input, while leaving important questions about curation, provenance and maintenance unresolved.

Digital art showing product designs transforming into structured text prompts on a screen, driven by a central luminous core bearing the Kage logo.

A pseudonymous maker using the name bustylasercanon introduced Kage on September 11th, giving developers a free library that turns designs from shipped websites into prompts for Claude Code, Codex and Cursor. The same maker uses the alias bustyLaserCannon on Reddit. No legal identity, employer or formal company role has been established.

Kage addresses a recurring problem in AI-assisted development: coding agents can produce an interface quickly, but the instruction "make it look good" supplies little usable visual direction. Kage packages references from existing products so developers can give an agent a more specific starting point.

From design reference to agent prompt

Kage lets users browse interfaces from real products and turn a selected design into a prompt. It also accepts website submissions, which the site says it captures, analyzes and adds to the library after review.

On September 11th, Kage's homepage displayed 160 designs, 780 components, 88 products, 17 skills and seven tools, alongside a site-reported count of 24 users online. Those figures are self-reported and unaudited.

The collection spans products in developer tools, AI, fintech, productivity, analytics and design. Kage also groups references into collections based on visual styles and technologies, including minimal and editorial landing pages, Tailwind CSS, Next.js and Vercel. The homepage presents individual references with a product name, page type, industry and visual descriptors.

That structure makes Kage closer to a working reference catalog than a gallery meant only for browsing. A developer can choose an existing interface, inspect how Kage classifies it and generate material for a coding agent. The site's positioning emphasizes design inspiration and prototyping rather than exact reproduction.

Kage's use of existing product designs also creates an unresolved provenance issue. The public materials available in the reporting record do not explain how captured designs are licensed or attributed, or how references are kept current as the original websites change. The submission review described on the homepage gives the maker a moderation role without disclosing the standards applied during that review.

A small catalog beside a much larger incumbent

Kage enters an established market for design references connected to coding agents. Mobbin's Model Context Protocol product describes a paid design-reference service with a large library of product screens and access for agents. Kage reports 160 designs and offers a simpler, free workflow centered on browsing a reference and turning it into a prompt. The two services count different units, so their headline catalog figures are not directly comparable.

Kage has not disclosed a legal entity, headquarters, funding, revenue or formal team structure. Its public launch is attached to the maker's aliases rather than a verified biography. That leaves the product itself as the evidence for the founder's approach: collect real interfaces, organize them into usable references and reduce the amount of visual interpretation a developer must express in a text prompt.

Free access may appeal to developers who need occasional design direction without another subscription. Maintaining that utility will require continued curation. Each submitted site adds work around selection, classification, provenance and updates, while a growing catalog risks becoming difficult to navigate if its labels and collections do not remain consistent.

For now, Kage offers a concrete input format for developers who already use coding agents: select a visual reference and pass the resulting prompt into the tool that will implement it. Its value will depend on the quality of those references and the clarity of the instructions Kage derives from them.

Reader comments

Conversation for this story loads after sign-in.